We present a Human Body model based IDentification system (HMID) system that is jointly trained for shape, pose and biometric identification. HMID is based on the Human Mesh Recovery (HMR) network and we propose additional losses to improve and stabilize shape estimation and biometric identification while maintaining the pose and shape output. We show that when our HMID network is trained using additional shape and pose losses, it shows a significant improvement in biometric identification performance when compared to an identical model that does not use such losses. The HMID model uses raw images instead of silhouettes and is able to perform robust recognition on images collected at range and altitude as many anthropometric properties are reasonably invariant to clothing, view and range. We show results on the USF dataset as well as the BRIAR dataset which includes probes with both clothing and view changes. Our approach (using body model losses) shows a significant improvement in Rank20 accuracy and True Accuracy Rate on the BRIAR evaluation dataset.

This paper addresses the problem of temporal pruning of noisy parts to improve event recognition performance. We present a new technique based on the temporal partitioning of the processed videos according to their motion patterns and the subsequent analysis of the yielded time segments. For each event type, we automatically learn the types of segments that are discriminative and those that perturb the classification. This process does not require detailed annotation of actions within an event type. A video is described with a set of quantized features and the final classification is performed according to the features that fall within the discriminative segments only. Experimental results show increased classification performance on the NIST MED11 dataset using two types of local features.

This paper describes a system for multimedia event detection and recounting. The goal is to detect a high level event class in unconstrained web videos and generate event oriented summarization for display to users. For this purpose, we detect informative segments and collect observations for them, leading to our ISOMER system. We combine a large collection of both low level and semantic level visual and audio features for event detection. For event recounting, we propose a novel approach to identify event oriented discriminative video segments and their descriptions with a linear SVM event classifier. User friendly concepts including objects, actions, scenes, speech and optical character recognition are used in generating descriptions. We also develop several mapping and filtering strategies to cope with noisy concept detectors. Our system performed competitively in the TRECVID 2013 Multimedia Event Detection task with near 100,000 videos and was the highest performer in TRECVID 2013 Multimedia Event Recounting task.

Analysis of massive track datasets is a challenging problem, especially when examining n-way relations inherent in social networks. In this paper, we use the Mitsubishi track database to examine the usefulness of three types of interaction features observable in tracklet networks. We explore ways in which social network information can be extracted and visualized using a statistical sampling of these features from a very large track dataset, with very little ground truth or outside knowledge. Special attention is given to methods that are likely to scale well beyond the size of the Mitsubishi dataset.

This study combines two useful methods in recognition: consensus or voting-based approaches and moment-based representations. Matches between image patches are generated using a Gaussian-weighted moment encoding of the patches and a feature indexing process. Each match implies an object 3D position and orientation (pose) and generates a vote for this pose. Recognition is accomplished by detecting significant clusters of votes in pose space. This combined method is an improvement over voting and moment methods in isolation. Using image brightness moments, the idea is demonstrated on examples of human faces undergoing full 3D pose change, as well as changes in features such as talking and blinking. The idea is then extended to moments of local texture orientation and successfully demonstrated under large variations in lighting nature and geometry.

The variation, with respect to view, of 2D features defined for projections of 3D point sets and line segments is studied. It is established that general-case view-invariants do not exist for any number of points, given true perspective, weak perspective, or orthographic projection models. Feature variation under the weak perspective approximation is then addressed. Though there are no general-case weak-perspective invariants, there are special-case invariants of practical importance. Those cited in the literature are derived from linear dependence relations and the invariance of this type of relation to linear transformation. The variation with respect to view is studied for an important set of 2D line segment features: the relative orientation, size, and position of one line segment with respect to another. The analysis includes an evaluation criterion for feature utility in terms of view-variation. This relationship is a function of both the feature and the particular configuration of 3D line segments. The use of this information in objection recognition is demonstrated for difficult discrimination tasks.>

The effective matching of a single 2D image of a cluttered scene to a library of multiple polyhedral models is achieved by organizing the 3D models into a network of descriptions of their 2D projections from expected views. The process of efficiently searching for image-model matches via a view description network is presented and demonstrated on images containing multiple objects and outdoor scenes. The experiments show that a recognition system based on view description networks is capable finding the correct matches to 3D objects in complex images with a potentially high level of efficiency.>

A method for planning smooth robot paths is presented. The method relies on the use of Laplace's equation to constrain the generation of a potential function over regions of the configuration space of an effector. Once the function is computed, paths may be found very quickly. These functions do not exhibit the local minima which plague the potential field method. Unlike decompositional and algebraic techniques. Laplace's equation is very well suited to computation on massively parallel architectures.>

This paper presents a new approach to the extraction of straight lines in intensity images. Pixels are grouped into line-support regions of similar gradient orientation, and then the structure of the associated intensity surface is used to determine the location and properties of the edge. The resulting regions and extracted edge parameters form a low-level representation of the intensity variations in the image that can be used for a variety of purposes. The algorithm appears to be more effective than previous techniques for two key reasons: 1) the gradient orientation (rather than gradient magnitude) is used as the initial organizing criterion prior to the extraction of straight lines, and 2) the global context of the intensity variations associated with a straight line is determined prior to any local decisions about participating edge elements.