Causal discovery, i.e., inferring underlying causal relationships from observational data, is highly challenging for AI systems. In a time series modeling context, traditional causal discovery methods mainly consider constrained scenarios with fully observed variables and/or data from stationary time-series. We develop a causal discovery approach to handle a wide class of nonstationary time series that are _conditionally stationary_, where the nonstationary behaviour is modeled as stationarity conditioned on a set of latent state variables. Named State-Dependent Causal Inference (SDCI), our approach is able to recover the underlying causal dependencies, with provable identifiablity for the state-dependent causal structures. Empirical experiments on nonlinear particle interaction data and gene regulatory networks demonstrate SDCI's superior performance over baseline causal discovery methods. Improved results over non-causal RNNs on modeling NBA player movements demonstrate the potential of our method and motivate the use of causality-driven methods for forecasting.

Data-driven three-dimensional parametric shape mod-els of the human body have gained enormous popularity both for the analysis of visual data and for the generation of synthetic humans. Following a similar approach for animals does not scale to the multitude of existing ani-mal species, not to mention the difficulty of accessing sub-jects to scan in 3D. However, we argue that for domestic species of great importance, like the horse, it is a highly valuable investment to put effort into gathering a large dataset of real 3D scans, and learn a realistic 3D articu-lated shape model. We introduce VAREN, a novel 3D ar-ticulated parametric shape model learned from 3D scans of many real horses. VAREN bridges synthesis and analysis tasks, as the generated model instances have unprecedented realism, while being able to represent horses of different sizes and shapes. Differently from previous body models, VAREN has two resolutions, an anatomical skeleton, and interpretable, learned pose-dependent deformations, which are related to the body muscles. We show with experiments that this formulation has superior performance with respect to previous strategies for modeling pose-dependent deformations in the human body case, while also being more compact and allowing an analysis of the relationship be-tween articulation and muscle deformation during articu-lated motion. The VAREN model and data are available at https://varen.is.tue.mpg.de.

The goal of Online Domain Adaptation for semantic segmentation is to handle unforeseeable domain changes that occur during deployment, like sudden weather events. However, the high computational costs associated with brute-force adaptation make this paradigm unfeasible for real-world applications. In this paper we propose HAMLET, a Hardware-Aware Modular Least Expensive Training framework for real-time domain adaptation. Our approach includes a hardware-aware back-propagation orchestration agent (HAMT) and a dedicated domain-shift detector that enables active control over when and how the model is adapted (LT). Thanks to these advancements, our approach is capable of performing semantic segmentation while simultaneously adapting at more than 29FPS on a single consumer-grade GPU. Our framework’s encouraging accuracy and speed trade-off is demonstrated on OnDA and SHIFT benchmarks through experimental results.

Data-driven and controllable human motion synthesis and prediction are active research areas with various applications in interactive media and social robotics. Challenges remain in these fields for generating diverse motions given past observations and dealing with imperfect poses. This paper introduces MoDiff, an autoregressive probabilistic diffusion model over motion sequences conditioned on control contexts of other modalities. Our model integrates a cross-modal Transformer encoder and a Transformer-based decoder, which are found effective in capturing temporal correlations in motion and control modalities. We also introduce a new data dropout method based on the diffusion forward process to provide richer data representations and robust generation. We demonstrate the superior performance of MoDiff in controllable motion synthesis for locomotion with respect to two baselines and show the benefits of diffusion data dropout for robust synthesis and reconstruction of high-fidelity motion close to recorded data.

Most action recognition models today are highly parameterized, and evaluated on datasets with appearance-wise distinct classes. It has also been shown that 2D Convolutional Neural Networks (CNNs) tend to be biased toward texture rather than shape in still image recognition tasks [19], in contrast to humans. Taken together, this raises suspicion that large video models partly learn spurious spatial texture correlations rather than to track relevant shapes over time to infer generalizable semantics from their movement. A natural way to avoid parameter explosion when learning visual patterns over time is to make use of recurrence. Biological vision consists of abundant recurrent circuitry, and is superior to computer vision in terms of domain shift generalization. In this article, we empirically study whether the choice of low-level temporal modeling has consequences for texture bias and cross-domain robustness. In order to enable a light-weight and systematic assessment of the ability to capture temporal structure, not revealed from single frames, we provide the Temporal Shape (TS) dataset, as well as modified domains of Diving48 allowing for the investigation of spatial texture bias in video models. The combined results of our experiments indicate that sound physical inductive bias such as recurrence in temporal modeling may be advantageous when robustness to domain shift is important for the task.

Abstract Advances in animal motion tracking and pose recognition have been a game changer in the study of animal behavior. Recently, an increasing number of works go ‘deeper’ than tracking, and address automated recognition of animals’ internal states such as emotions and pain with the aim of improving animal welfare, making this a timely moment for a systematization of the field. This paper provides a comprehensive survey of computer vision-based research on recognition of pain and emotional states in animals, addressing both facial and bodily behavior analysis. We summarize the efforts that have been presented so far within this topic—classifying them across different dimensions, highlight challenges and research gaps, and provide best practice recommendations for advancing the field, and some future directions for research.

Multi-task learning requires accurate identification of the correlations between tasks. In real-world time-series, tasks are rarely perfectly temporally aligned; traditional multi-task models do not account for this and subsequent errors in correlation estimation will result in poor predictive performance and uncertainty quantification. We introduce a method that automatically accounts for temporal misalignment in a unified generative model that improves predictive performance. Our method uses Gaussian processes (GPs) to model the correlations both within and between the tasks. Building on the previous work by Kazlauskaite et al. (2019), we include a separate monotonic warp of the input data to model temporal misalignment. In contrast to previous work, we formulate a lower bound that accounts for uncertainty in both the estimates of the warping process and the underlying functions. Also, our new take on a monotonic stochastic process, with efficient path-wise sampling for the warp functions, allows us to perform full Bayesian inference in the model rather than MAP estimates. Missing data experiments, on synthetic and real time-series, demonstrate the advantages of accounting for misalignments (vs standard unaligned method) as well as modelling the uncertainty in the warping process (vs baseline MAP alignment approach).

Timely detection of horse pain is important for equine welfare. Horses express pain through their facial and body behavior, but may hide signs of pain from unfamiliar human observers. In addition, collecting visual data with detailed annotation of horse behavior and pain state is both cumbersome and not scalable. Consequently, a pragmatic equine pain classification system would use video of the unobserved horse and weak labels. This paper proposes such a method for equine pain classification by using multi-view surveillance video footage of unobserved horses with induced orthopaedic pain, with temporally sparse video level pain labels. To ensure that pain is learned from horse body language alone, we first train a self-supervised generative model to disentangle horse pose from its appearance and background before using the disentangled horse pose latent representation for pain classification. To make best use of the pain labels, we develop a novel loss that formulates pain classification as a multi-instance learning problem. Our method achieves pain classification accuracy better than human expert performance with 60% accuracy. The learned latent horse pose representation is shown to be viewpoint covariant, and disentangled from horse appearance. Qualitative analysis of pain classified segments shows correspondence between the pain symptoms identified by our model, and equine pain scales used in veterinary practice.

Missing values with mixed data types is a common problem in a large number of machine learning applications such as processing of surveys and in different medical applications. Recently, Gaussian copula models have been suggested as a means of performing imputation of missing values using a probabilistic framework. While the present Gaussian copula models have shown to yield state of the art performance, they have two limitations: they are based on an approximation that is fast but may be imprecise and they do not support unordered multinomial variables. We address the first limitation using direct and arbitrarily precise approximations both for model estimation and imputation by using randomized quasi-Monte Carlo procedures. The method we provide has lower errors for the estimated model parameters and the imputed values, compared to previously proposed methods. We also extend the previous Gaussian copula models to include unordered multinomial variables in addition to the present support of ordinal, binary, and continuous variables.

We propose a new framework for gesture generation, aiming to allow data-driven approaches to produce more semantically rich gestures. Our approach first predicts whether to gesture, followed by a prediction of the gesture properties. Those properties are then used as conditioning for a modern probabilistic gesture-generation model capable of high-quality output. This empowers the approach to generate gestures that are both diverse and representational. Follow-ups and more information can be found on the project page: https://svito-zar.github.io/speech2properties2gestures/

This paper presents a novel framework for speech-driven gesture production, applicable to virtual agents to enhance human-computer interaction. Specifically, we extend recent deep-learning-based, data-driven methods for speech-driven gesture generation by incorporating representation learning. Our model takes speech as input and produces gestures as output, in the form of a sequence of 3D coordinates. We provide an analysis of different representations for the input (speech) and the output (motion) of the network by both objective and subjective evaluations. We also analyze the importance of smoothing of the produced motion. Our results indicated that the proposed method improved on our baseline in terms of objective measures. For example, it better captured the motion dynamics and better matched the motion-speed distribution. Moreover, we performed user studies on two different datasets. The studies confirmed that our proposed method is perceived as more natural than the baseline, although the difference in the studies was eliminated by appropriate post-processing: hip-centering and smoothing. We conclude that it is important to take both motion representation and post-processing into account when designing an automatic gesture-production method.

Many applications of classification methods not only require high accuracy but also reliable estimation of predictive uncertainty. However, while many current classification frameworks, in particular deep neural networks, achieve high accuracy, they tend to incorrectly estimate uncertainty. In this paper, we propose a method that adjusts the confidence estimates of a general classifier such that they approach the probability of classifying correctly. In contrast to existing approaches, our calibration method employs a non-parametric representation using a latent Gaussian process, and is specifically designed for multi-class classification. It can be applied to any classifier that outputs confidence estimates and is not limited to neural networks. We also provide a theoretical analysis regarding the over- and underconfidence of a classifier and its relationship to calibration, as well as an empirical outlook for calibrated active learning. In experiments we show the universally strong performance of our method across different classifiers and benchmark data sets, in particular for state-of-the art neural network architectures.

During speech, people spontaneously gesticulate, which plays a key role in conveying information. Similarly, realistic co-speech gestures are crucial to enable natural and smooth interactions with social agents. Current end-to-end co-speech gesture generation systems use a single modality for representing speech: either audio or text. These systems are therefore confined to producing either acoustically-linked beat gestures or semantically-linked gesticulation (e.g., raising a hand when saying "high''): they cannot appropriately learn to generate both gesture types. We present a model designed to produce arbitrary beat and semantic gestures together. Our deep-learning based model takes both acoustic and semantic representations of speech as input, and generates gestures as a sequence of joint angle rotations as output. The resulting gestures can be applied to both virtual agents and humanoid robots. Subjective and objective evaluations confirm the success of our approach. The code and video are available at the project page svito-zar.github.io/gesticulator .

Auto encoding models have been extensively studied in recent years. They provide an efficient framework for sample generation, as well as for analysing feature learning. Furthermore, they are efficient in performing interpolations between data-points in semantically meaningful ways. In this paper, we introduce a method for generating sequence samples from auto encoders trained on flattened sequences (e.g video sample from auto encoders trained on single frames); as well as a canonical, dimension independent method for generating stochastic interpolations. The distribution of interpolation paths is represented as the distribution of a bridge process constructed from an artificial random data generating process in the latent space, having the prior distribution as its invariant distribution.

Scene understanding is paramount in robotics, self-navigation, augmented reality, and many other fields. To fully accomplish this task, an autonomous agent has to infer the 3D structure of the sensed scene (to know where it looks at) and its content (to know what it sees). To tackle the two tasks, deep neural networks trained to infer semantic segmentation and depth from stereo images are often the preferred choices. Specifically, Semantic Stereo Matching can be tackled by either standalone models trained for the two tasks independently or joint end-to-end architectures. Nonetheless, as proposed so far, both solutions are inefficient because requiring two forward passes in the former case or due to the complexity of a single network in the latter, although jointly tackling both tasks is usually beneficial in terms of accuracy. In this paper, we propose a single compact and lightweight architecture for real-time semantic stereo matching. Our framework relies on coarse-to-fine estimations in a multi-stage fashion, allowing: i) very fast inference even on embedded devices, with marginal drops in accuracy, compared to state-of-the-art networks, ii) trade accuracy for speed, according to the specific application requirements. Experimental results on high-end GPUs as well as on an embedded Jetson TX2 confirm the superiority of semantic stereo matching compared to standalone tasks and highlight the versatility of our framework on any hardware and for any application.

Although many fairness criteria have been proposed for decision making, their long-term impact on the well-being of a population remains unclear. In this work, we study the dynamics of population qualification and algorithmic decisions under a partially observed Markov decision problem setting. By characterizing the equilibrium of such dynamics, we analyze the long-term impact of static fairness constraints on the equality and improvement of group well-being. Our results show that static fairness constraints can either promote equality or exacerbate disparity depending on the driving factor of qualification transitions and the effect of sensitive attributes on feature distributions. We also consider possible interventions that can effectively improve group qualification or promote equality of group qualification. Our theoretical results and experiments on static real-world datasets with simulated dynamics show that our framework can be used to facilitate social science studies.

We present a method for weakly-supervised action localization based on graph convolutions. In order to find and classify video time segments that correspond to relevant action classes, a system must be able to both identify discriminative time segments in each video, and identify the full extent of each action. Achieving this with weak video level labels requires the system to use similarity and dissimilarity between moments across videos in the training data to understand both how an action appears, as well as the subactions that comprise the action's full extent. However, current methods do not make explicit use of similarity between video moments to inform the localization and classification predictions. We present a novel method that uses graph convolutions to explicitly model similarity between video moments. Our method utilizes similarity graphs that encode appearance and motion, and pushes the state of the art on THUMOS'14, ActivityNet 1.2, and Charades for weakly- supervised action localization.

Missing data are ubiquitous in many domains such as healthcare. When these data entries are not missing completely at random, the (conditional) independence relations in the observed data may be different from those in the complete data generated by the underlying causal process. Consequently, simply applying existing causal discovery methods to the observed data may lead to wrong conclusions. In this paper, we aim at developing a causal discovery method to recover the underlying causal structure from observed data that are missing under different mechanisms, including missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR). With missingness mechanisms represented by missingness graphs (m-graphs), we analyze conditions under which additional correction is needed to derive conditional independence/dependence relations in the complete data. Based on our analysis, we propose Missing Value PC (MVPC), which extends the PC algorithm to incorporate additional corrections. Our proposed MVPC is shown in theory to give asymptotically correct results even on data that are MAR or MNAR. Experimental results on both synthetic data and real healthcare applications illustrate that the proposed algorithm is able to find correct causal relations even in the general case of MNAR.

Movement-based interactions are gaining traction, requiring a better understanding of how such expressions are shaped by designers. Through an analysis of an artistic process aimed to deliver a commissioned opera where custom-built drones are performing on stage alongside human performers, we observed the importance of achieving an intercorporeal understanding to shape body-based emotional expressivity. Our analysis reveals how the choreographer moves herself to: (1) imitate and feel the affordances and expressivity of the drones' 'otherness' through her own bodily experience; (2) communicate to the engineer of the team how she wants to alter the drones' behaviors to be more expressive; (3) enact and interactively alter her choreography. Through months of intense development and creative work, such an intercorporeal understanding was achieved by carefully crafting the drones' behaviors, but also by the choreographer adjusting her own somatics and expressions. The choreography arose as a result of the expressivity they enabled together.

A prerequisite to successfully alleviate pain in animals is to recognize it, which is a great challenge in non-verbal species. Furthermore, prey animals such as horses tend to hide their pain. In this study, we propose a deep recurrent two-stream architecture for the task of distinguishing pain from non-pain in videos of horses. Different models are evaluated on a unique dataset showing horses under controlled trials with moderate pain induction, which has been presented in earlier work. Sequential models are experimentally compared to single-frame models, showing the importance of the temporal dimension of the data, and are benchmarked against a veterinary expert classification of the data. We additionally perform baseline comparisons with generalized versions of state-of-the-art human pain recognition methods. While equine pain detection in machine learning is a novel field, our results surpass veterinary expert performance and outperform pain detection results reported for other larger non-human species.

This paper presents a novel framework for automatic speech-driven gesture generation, applicable to human-agent interaction including both virtual agents and robots. Specifically, we extend recent deep-learning-based, data-driven methods for speech-driven gesture generation by incorporating representation learning. Our model takes speech as input and produces gestures as output, in the form of a sequence of 3D coordinates.

Discovery of causal relations from observational data is essential for many disciplines of science and real-world applications. However, unlike other machine learning algorithms, whose development has been greatly fostered by a large amount of available benchmark datasets, causal discovery algorithms are notoriously difficult to be systematically evaluated because few datasets with known ground-truth causal relations are available. In this work, we handle the problem of evaluating causal discovery algorithms by building a flexible simulator in the medical setting. We develop a neuropathic pain diagnosis simulator, inspired by the fact that the biological processes of neuropathic pathophysiology are well studied with well-understood causal influences. Our simulator exploits the causal graph of the neuropathic pain pathology and its parameters in the generator are estimated from real-life patient cases. We show that the data generated from our simulator have similar statistics as real-world data. As a clear advantage, the simulator can produce infinite samples without jeopardizing the privacy of real-world patients. Our simulator provides a natural tool for evaluating various types of causal discovery algorithms, including those to deal with practical issues in causal discovery, such as unknown confounders, selection bias, and missing data. Using our simulator, we have evaluated extensively causal discovery algorithms under various settings.

Image classification models built into visual support systems and other assistive devices need to provide accurate predictions about their environment. We focus on an application of assistive technology for people with visual impairments, for daily activities such as shopping or cooking. In this paper, we provide a new benchmark dataset for a challenging task in this application - classification of fruits, vegetables, and refrigerated products, e.g. milk packages and juice cartons, in grocery stores. To enable the learning process to utilize multiple sources of structured information, this dataset not only contains a large volume of natural images but also includes the corresponding information of the product from an online shopping website. Such information encompasses the hierarchical structure of the object classes, as well as an iconic image of each type of object. This dataset can be used to train and evaluate image classification models for helping visually impaired people in natural environments. Additionally, we provide benchmark results evaluated on pretrained convolutional neural networks often used for image understanding purposes, and also a multi-view variational autoencoder, which is capable of utilizing the rich product information in the dataset.

Many modern unsupervised or semi-supervised machine learning algorithms rely on Bayesian probabilistic models. These models are usually intractable and thus require approximate inference. Variational inference (VI) lets us approximate a high-dimensional Bayesian posterior with a simpler variational distribution by solving an optimization problem. This approach has been successfully applied to various models and large-scale applications. In this review, we give an overview of recent trends in variational inference. We first introduce standard mean field variational inference, then review recent advances focusing on the following aspects: (a) scalable VI, which includes stochastic approximations, (b) generic VI, which extends the applicability of VI to a large class of otherwise intractable models, such as non-conjugate models, (c) accurate VI, which includes variational models beyond the mean field approximation or with atypical divergences, and (d) amortized VI, which implements the inference over local latent variables with inference networks. Finally, we provide a summary of promising future research directions.

This study proposes, develops, and evaluates methods for modeling the eye-gaze direction and head orientation of a person in multiparty open-world dialogues, as a function of low-level communicative signals generated by his/hers interlocutors. These signals include speech activity, eye-gaze direction, and head orientation, all of which can be estimated in real time during the interaction. By utilizing these signals and novel data representations suitable for the task and context, the developed methods can generate plausible candidate gaze targets in real time. The methods are based on Feedforward Neural Networks and Long Short-Term Memory Networks. The proposed methods are developed using several hours of unrestricted interaction data and their performance is compared with a heuristic baseline method. The study offers an extensive evaluation of the proposed methods that investigates the contribution of different predictors to the accurate generation of candidate gaze targets. The results show that the methods can accurately generate candidate gaze targets when the person being modeled is in a listening state. However, when the person being modeled is in a speaking state, the proposed methods yield significantly lower performance.

Dissimilarity measures are often used as a proxy or a handle to reason about data. This can be problematic, as the data representation is often a consequence of the capturing process or how the data is visualized, rather than a reflection of the semantics that we want to extract. Facial expressions are a subtle and essential part of human communication but they are challenging to extract from current representations. In this paper we present a method that is capable of learning semantic representations of faces in a data driven manner. Our approach uses sparse human supervision which our method grounds in the data. We provide experimental justification of our approach showing that our representation improves the performance for emotion classification.

Fluent and safe interactions of humans and robots require both partners to anticipate the others' actions. The bottleneck of most methods is the lack of an accurate model of natural human motion. In this work, we present a conditional variational autoencoder that is trained to predict a window of future human motion given a window of past frames. Using skeletal data obtained from RGB depth images, we show how this unsupervised approach can be used for online motion prediction for up to 1660 ms. Additionally, we demonstrate online target prediction within the first 300-500 ms after motion onset without the use of target specific training data. The advantage of our probabilistic approach is the possibility to draw samples of possible future motion patterns. Finally, we investigate how movements and kinematic cues are represented on the learned low dimensional manifold.

Generative models of 3D human motion are often restricted to a small number of activities and can therefore not generalize well to novel movements or applications. In this work we propose a deep learning framework for human motion capture data that learns a generic representation from a large corpus of motion capture data and generalizes well to new, unseen, motions. Using an encoding-decoding network that learns to predict future 3D poses from the most recent past, we extract a feature representation of human motion. Most work on deep learning for sequence prediction focuses on video and speech. Since skeletal data has a different structure, we present and evaluate different network architectures that make different assumptions about time dependencies and limb correlations. To quantify the learned features, we use the output of different layers for action classification and visualize the receptive fields of the network units. Our method outperforms the recent state of the art in skeletal motion prediction even though these use action specific training data. Our results show that deep feedforward networks, trained from a generic mocap database, can successfully be used for feature extraction from human motion data and that this representation can be used as a foundation for classification and prediction.

This paper presents a unique dataset consisting of 20 recordings of the same musical piece, conducted with 4 different musical intentions in mind. The upper body and baton motion of a professional conductor was recorded, as well as the sound of each instrument in a professional string quartet following the conductor. The dataset is made available for benchmarking of motion recognition algorithms. An HMM-based emotion intent classification method is trained with subsets of the data, and classification of other subsets of the data show firstly that the motion of the baton communicates energetic intention to a high degree, secondly, that the conductor’s torso, head and other arm conveys calm intention to a high degree, and thirdly, that positive vs negative sentiments are communicated to a high degree through other channels than the body and baton motion – most probably, through facial expression and muscle tension conveyed through articulated hand and finger motion. The long-term goal of this work is to develop a computer model of the entire conductor-orchestra communication process; the studies presented here indicate that computer modeling of the conductor-orchestra communication is feasible.

While many advances towards effective anomaly detection techniques targeting specific applications have been made in recent years, little work has been done to develop application-agnostic approaches to the subject. In this article, we present such an approach, in which anomaly detection methods are treated as formal, structured objects. We consider a general class of methods, with an emphasis on methods that utilize structural properties of the data they operate on. For this class of methods, we develop a decomposition into sub-methods—simple, restricted objects, which may be reasoned about independently and combined to form methods. As we show, this formalism enables the construction of software that facilitates formulating, implementing, evaluating, as well as algorithmically finding and calibrating anomaly detection methods.

This work employs an adaptive learning mechanism to perform tracking of an unknown object through RGBD cameras. We extend our previous framework to robustly track a wider range of arbitrarily shaped objects by adapting the model to the measured object size. The size is estimated as the object undergoes motion, which is done by fitting an inscribed cuboid to the measurements. The region spanned by this cuboid is used during tracking, to determine whether or not new measurements should be added to the object model. In our experiments we test our tracker with a set of objects of arbitrary shape and we show the benefit of the proposed model due to its ability to adapt to the object shape which leads to more robust tracking results.

Visual tracking of unknown objects is an essential task in robotic perception, of importance to a wide range of applications. In the general scenario, the robot has no full 3D model of the object beforehand, just the partial view of the object visible in the first video frame. A tracker with this information only will inevitably lose track of the object after occlusions or large out-of-plane rotations. The way to overcome this is to incrementally learn the appearances of new views of the object. However, this bootstrapping approach is sensitive to drifting due to occasional inclusion of the background into the model.

Forensic video analysis is the offline analysis of video aimed at understanding what happened in a scene in the past. Two of its key tasks are the recognition of specific actions, e.g., walking or fighting, and the search for specific persons, also referred to as re-identification. Although these tasks have traditionally been performed manually in forensic investigations, the current growing number of cameras and recorded video leads to the need for automated analysis. In this paper we propose an unsupervised retrieval system for surveillance videos based on human action and appearance. Given a query window, the system retrieves people performing the same action as the one in the query, the same person performing any action, or the same person performing the same action. We use an adaptive search algorithm that focuses the analysis on relevant frames based on the inter-frame difference of foreground masks. Then, for each analyzed frame, a pedestrian detector is used to extract windows containing each pedestrian in the scene. For each detection, we use optical flow features to represent its action and color features to represent its appearance. These extracted features are used to compute the probability that the detection matches the query according to the specified criterion. The algorithm is fully unsupervised, i.e., no training or constraints on the appearance, actions or number of actions that will appear in the test video are made. The proposed algorithm is tested on a surveillance video with different people performing different actions, providing satisfactory retrieval performance.

Humans are able to merge information from multiple perceptional modalities and formulate a coherent representation of the world. Our thesis is that robots need to do the same in order to operate robustly and autonomously in an unstructured environment. It has also been shown in several fields that multiple sources of information can complement each other, overcoming the limitations of a single perceptual modality. Hence, in this paper we introduce a data set of actions that includes both visual data (RGB-D video and 6DOF object pose estimation) and acoustic data. We also propose a method for recognizing and segmenting actions from continuous audio-visual data. The proposed method is employed for extensive evaluation of the descriptive power of the two modalities, and we discuss how they can be used jointly to infer a coherent interpretation of the recorded action.

In order for robots to function in unstructured environments in interaction with humans, they must be able to reason about the world in a semantic meaningful way. An essential capability is to segment the world into semantic plausible object hypotheses. In this paper we propose a general framework which can be used for reasoning about objects and their functionality in manipulation activities. Our system employs a hierarchical segmentation framework that extracts object hypotheses from RGB-D video. Motivated by cognitive studies on humans, our work leverages on contextual information, e.g., that objects obey the laws of physics, to formulate object hypotheses from regions in a mathematically principled manner.

We propose a new approach for learning a summarized representation of high dimensional continuous data. Our technique consists of a Bayesian non-parametric model capable of encoding high-dimensional data from complex distributions using a sparse summarization. Specifically, the method marries techniques from probabilistic dimensionality reduction and clustering. We apply the model to learn efficient representations of grasping data for two robotic scenarios.

The ability to learn from human demonstration is essential for robots in human environments. The activity models that the robot builds from observation must take both the human motion and the objects involved into account. Object models designed for this purpose should reflect the role of the object in the activity - its function, or affordances. The main contribution of this paper is to represent object directly in terms of their interaction with human hands, rather than in terms of appearance. This enables the direct representation of object affordances/function, while being robust to intra-class differences in appearance. Object hypotheses are first extracted from a video sequence as tracks of associated image segments. The object hypotheses are encoded as strings, where the vocabulary corresponds to different types of interaction with human hands. The similarity between two such object descriptors can be measured using a string kernel. Experiments show these functional descriptors to capture differences and similarities in object affordances/function that are not represented by appearance.

Learning about activities and object affordances from human demonstration are important cognitive capabilities for robots functioning in human environments, for example, being able to classify objects and knowing how to grasp them for different tasks. To achieve such capabilities, we propose a Labeled Multi-modal Latent Dirichlet Allocation (LM-LDA), which is a generative classifier trained with two different data cues, for instance, one cue can be traditional visual observation and another cue can be contextual information. The novel aspects of the LM-LDA classifier, compared to other methods for encoding contextual information are that, I) even with only one of the cues present at execution time, the classification will be better than single cue classification since cue correlations are encoded in the model, II) one of the cues (e.g., common grasps for the observed object class) can be inferred from the other cue (e.g., the appearance of the observed object). This makes the method suitable for robot online and transfer learning; a capability highly desirable in cognitive robotic applications. Our experiments show a clear improvement for classification and a reasonable inference of the missing data.

We address the problem of representing and encoding human hand motion data using nonlinear dimensionality reduction methods. We build our work on the notion of postural synergies being typically based on a linear embedding of the data. In addition to addressing the encoding of postural synergies using nonlinear methods, we relate our work to control strategies of combined reaching and grasping movements. We show the drawbacks of the (commonly made) causality assumption and propose methods that model the data as being generated from an inferred latent manifold to cope with the problem. Another important contribution is a thorough analysis of the parameters used in the employed dimensionality reduction techniques. Finally, we provide an experimental evaluation that shows how the proposed methods outperform the standard techniques, both in terms of recognition and generation of motion patterns.

We present a novel method for learning densities with bounded support which enables us to incorporate `hard' topological constraints. In particular, we show how emerging techniques from computational algebraic topology and the notion of Persistent Homology can be combined with kernel based methods from Machine Learning for the purpose of density estimation. The proposed formalism facilitates learning of models with bounded support in a principled way, and -- by incorporating Persistent Homology techniques in our approach -- we are able to encode algebraic-topological constraints which are not addressed in current state-of the art probabilistic models. We study the behaviour of our method on two synthetic examples for various sample sizes and exemplify the benefits of the proposed approach on a real-world data-set by learning a motion model for a racecar. We show how to learn a model which respects the underlying topological structure of the racetrack, constraining the trajectories of the car.

While the problem of tracking 3D human motion has been widely studied, most approaches have assumed that the person is isolated and not interacting with the environment. Environmental constraints, however, can greatly constrain and simplify the tracking problem. The most studied constraints involve gravity and contact with the ground plane. We go further to consider interaction with objects in the environment. In many cases, tracking rigid environmental objects is simpler than tracking high-dimensional human motion. When a human is in contact with objects in the world, their poses constrain the pose of body, essentially removing degrees of freedom. Thus what would appear to be a harder problem, combining object and human tracking, is actually simpler. We use a standard formulation of the body tracking problem but add an explicit model of contact with objects. We find that constraints from the world make it possible to track complex articulated human motion in 3D from a monocular camera.

This paper presents a method for vision based estimation of the pose of human hands in interaction with objects. Despite the fact that most robotics applications of human hand tracking involve grasping and manipulation of objects, the majority of methods in the literature assume a free hand, isolated from the surrounding environment. Our hand tracking method is non-parametric, performing a nearest neighbor search in a large database (100000 entries) of hand poses with and without grasped objects. The system operates in real time, it is robust to self occlusions, object occlusions and segmentation errors, and provides full hand pose reconstruction from markerless video. Temporal consistency in hand pose is taken into account, without explicitly tracking the hand in the high dimensional pose space.

Understanding the spatial dimensionality and temporal context of human hand actions can provide representations for programming grasping actions in robots and inspire design of new robotic and prosthetic hands. The natural representation of human hand motion has high dimensionality. For specific activities such as handling and grasping of objects, the commonly observed hand motions lie on a lower-dimensional non-linear manifold in hand posture space. Although full body human motion is well studied within Computer Vision and Biomechanics, there is very little work on the analysis of hand motion with nonlinear dimensionality reduction techniques. In this paper we use Gaussian Process Latent Variable Models (GPLVMs) to model the lower dimensional manifold of human hand motions during object grasping. We show how the technique can be used to embed high-dimensional grasping actions in a lower-dimensional space suitable for modeling, recognition and mapping.

This paper presents a vision based method for grasp classification. It is developed as part of a Programming by Demonstration (PbD) system for which recognition of objects and pick-and-place actions represent basic building blocks for task learning. In contrary to earlier approaches, no articulated 3D reconstruction of the hand over time is taking place. The indata consists of a single image of the human hand. A 2D representation of the hand shape, based on gradient orientation histograms, is extracted from the image. The hand shape is then classified as one of six grasps by finding similar hand shapes in a large database of grasp images. The database search is performed using Locality Sensitive Hashing (LSH), an approximate k-nearest neighbor approach. The nearest neighbors also give an estimated hand orientation with respect to the camera. The six human grasps are mapped to three Barret hand grasps. Depending on the type of robot grasp, a precomputed grasp strategy is selected. The strategy is further parameterized by the orientation of the hand relative to the object. To evaluate the potential for the method to be part of a robust vision system, experiments were performed, comparing classification results to a baseline of human classification performance. The experiments showed the LSH recognition performance to be comparable to human performance.

We present a method for audio-visual classification of Swedish phonemes, to be used in computer-assisted pronunciation training. The probabilistic kernel-based method is applied to the audio signal and/or either a principal or an independent component (PCA or ICA) representation of the mouth region in video images. We investigate which representation (PCA or ICA) that may be most suitable and the number of components required in the base, in order to be able to automatically detect pronunciation errors in Swedish from audio-visual input. Experiments performed on one speaker show that the visual information help avoiding classification errors that would lead to gravely erroneous feedback to the user; that it is better to perform phoneme classification on audio and video separately and then fuse the results, rather than combining them before classification; and that PCA outperforms ICA for fewer than 50 components. Index Terms: audiovisual phoneme classification, pronunciation error detection, PCA, ICA

This paper presents an approach to articulatory inversion using audio and video of the user’s face, requiring no special markers. The video is stabilized with respect to the face, and the mouth region cropped out. The mouth image is projected into a learned independent component subspace to obtain a low-dimensional representation of the mouth appearance. The inversion problem is treated as one of regression; a non-linear regressor using relevance vector machines is trained with a dataset of simultaneous images of a subject’s face, acoustic features and positions of magnetic coils glued to the subjects’s tongue. The results show the benefit of using both cues for inversion. We envisage the inversion method to be part of a pronunciation training system with articulatory feedback. Index Terms: audio-visual to articulatory inversion. 1.

This study has been performed in order to evaluate a prototype for the human – computer interface of a computer-based speech training aid named ARTUR. The main feature of the aid is that it can give suggestions on how to improve articulations. Two user groups were involved: three children aged 9 – 14 with extensive experience of speech training with therapists and computers, and three children aged 6, with little or no prior experience of computer-based speech training. All children had general language disorders. The study indicates that the present interface is usable without prior training or instructions, even for the younger children, but that more motivational factors should be introduced. The granularity of the mesh that classifies mispronunciations was satisfactory, but the flexibility and level of detail of the feedback should be developed further.

This study has been performed in order to test the human-machine interface of a computer-based speech training aid named ARTUR with the main feature that it can give suggestions on how to improve articulation. Two user groups were involved: three children aged 9-14 with extensive experience of speech training, and three children aged 6. All children had general language disorders.The study indicates that the present interface is usable without prior training or instructions, even for the younger children, although it needs some improvement to fit illiterate children. The granularity of the mesh that classifies mispronunciations was satisfactory, but can be developed further.

This paper describes a framework for learning probabilistic models of objects and scenes and for exploiting these models for tracking complex, deformable, or articulated objects in image sequences. We focus on the probabilistic tracking of people and learn models of how they appear and move in images. In particular we learn the likelihood of observing various spatial and temporal filter responses corresponding to edges, ridges, and motion differences given a model of the person. Similarly, we learn probability distributions over filter responses for general scenes that define a likelihood of observing the filter responses for arbitrary backgrounds. We then derive a probabilistic model for tracking that exploits the ratio between the likelihood that image pixels corresponding to the foreground (person) were generated by an actual person or by some unknown background. The paper extends previous work on learning image statistics and combines it with Bayesian tracking using particle filtering. By combining multiple image cues, and by using learned likelihood models, we demonstrate improved robustness and accuracy when tracking complex objects such as people in monocular image sequences with cluttered scene and a moving camera.

This paper describes a framework for constructing a linear subspace model of image appearance for complex articulated 3D figures such as humans and other animals. A commercial motion capture system provides 3D data that is aligned with images of subjects performing various activities. Portions of a limb's image appearance are seen from multiple views and for multiple subjects. From these partial views, weighted principal component analysis is used to construct a linear subspace representation of the "unwrapped" image appearance of each limb. The linear subspaces provide a generative model of the object appearance that is exploited in a Bayesian particle filtering tracking system. Results of tracking single limbs and walking humans are presented.

We present methods for learning and tracking human motion in video. We estimate a statistical model of typical activities from a large set of 3D periodic human motion data by segmenting these data automatically into "cycles". Then the mean and the princi(cid:173) pal components of the cycles are computed using a new algorithm that accounts for missing information and enforces smooth tran(cid:173) sitions between cycles. The learned temporal model provides a prior probability distribution over human motions that can be used in a Bayesian framework for tracking human subjects in complex monocular video sequences and recovering their 3D motion.

In this paper, a person following behaviour for a mobile robot is presented. The head of the person is located using skin colour detection. Then, a control loop is fed with the camera movements required to put the upper part of the person in the center of the image. The algorithm was tested in different rooms of a research lab. It performed well in all lighting conditions except in direct sunlight. Since the background and lighting cannot be controlled, the vision algorithm must be robust to such changes. However, since the computing power is quite limited, the algorithm must have as low complexity as possible.