Abstract In augmented reality applications it is essential to know the position and orientation of the user to correctly register virtual 3D content in the user’s field of view. For this purpose, visual tracking through simultaneous localization and mapping (SLAM) is often used. However, when applied to the commonly occurring situation where the users are mostly stationary, many methods presented in previous research have two key limitations. First, SLAM techniques alone do not address the problem of global localization with respect to prior models of the environment. Global localization is essential in many applications where multiple users are expected to track within a shared space, such as spectators at a sporting event. Secondly, these methods often assume significant translational movement to accurately reconstruct and track from a local model of the environment, causing challenges for many stationary applications. In this paper, we extend recent research on Spherical Localization and Tracking to support relocalization after tracking failure, as well as global localization in large shared environments, and optimize the method for operation on mobile hardware. We also evaluate various state-of-the-art localization approaches, the robustness of our visual tracking method, and demonstrate the effectiveness of our system in real-life scenarios.

An electronic voting (E-voting) oriented cybersecurity curriculum, proposed by Hostler et al. [4] in 2021, leverages the rich security features of E-voting systems and E-voting process to teach essential concepts of cybersecurity. Existing curricular guidelines describe topics in computer security, but do not instantiate them with examples. This is because their goals are different. In this case study, we map the e-voting curriculum into the CSEC2017 curriculum guidelines, to demonstrate how such a mapping is done. Further, this enables teachers to select the parts of the e-voting curriculum most relevant to their classes, by basing the selection on the relevant CSEC2017 learning objectives. We conclude with a brief discussion on generalizing this mapping to other curricular guidelines.

As a result of urban concentration, colonisation, increased physical distance from tribal homelands, and globalisation, many indigenous people, including Māori, are seeking digital solutions to connect to their culture and identity. Through co-design, close collaboration with our indigenous partners, and careful consideration of cultural context, we show that mixed reality experiences can be an effective mechanism for the growing diaspora of Māori to access and experience their language, genealogy, families, histories and knowledge. Inductive analysis of semi-structured interviews highlights the importance of cultural values and context in this experience, and confirms that this approach can support connections to community and culture. Our work is deeply embedded in a particular indigenous group’s context and culture, but we believe that it holds important lessons that can generalise to other groups. In particular, collaborative co-design and recognition of cultural values throughout the project are essential for producing experiences that meet the needs of specific communities, and that reflect and respect their culture and worldview.

We introduce a deep recursive octree network for the compression of 3D voxel data. Our network compresses a voxel grid of any size down to a very small latent space in an autoencoder-like network. We show results for compressing 323, 643and 1283grids down to just 80 floats in the latent space. We demonstrate the effectiveness and efficiency of our proposed method on several publicly available datasets with three experiments: 3D shape classification, 3D shape reconstruction, and shape generation. Experimental results show that our algorithm maintains accuracy while consuming less memory with shorter training times compared to existing methods, especially in 3D reconstruction tasks.

Hand-held capture of stereo panoramas involves spinning the camera in a roughly circular path to acquire a dense set of views of the scene. However, most existing structure-from-motion pipelines fail when trying to reconstruct such trajectories, due to the small baseline between frames. In this work, we evaluate the use of spherical structure-from-motion for reconstructing handheld stereo panorama captures. The spherical motion constraint introduces a strong regularization on the structure-from-motion process which mitigates the small-baseline problem, making it well-suited to the use case of stereo panorama capture with a handheld camera. We demonstrate the effectiveness of spherical structure-from-motion for casual capture of high-resolution stereo panoramas and validate our results with a user study.

When implementing an Augmented Reality (AR) interface, it is essential to track camera motion in order to precisely register the virtual overlay in the view of the user. However, unlike most indoor AR scenarios, in many outdoor scenarios the user maintains a static position performing mostly rotational movements. Simultaneous Localization and Mapping (SLAM) methods typically used to solve the tracking problem require significant translational camera motion to perform reliably. The magnitude of the required translation is proportional to the size of the scene, exacerbating this problem in large environments such as open places or stadiums. In this paper, we present an alternative SLAM method, which combines spherical Structure-from-Motion and a robust 3D tracking method. We compare our method to ORB SLAM2 in synthetic and real tests, and show that our method can track more reliably in large spaces, with simpler calculation due to the spherical motion constraint. We discuss this issue in the context of implementing an AR interface for live sport events in stadiums or other open environments, but possible application scenarios for our technique go beyond and can be applied to handheld AR in many outdoor environments.

Augmented Reality (AR) sandtables facilitate the shaping of sand to form a surface that is transformed into a digital terrain map which is projected back onto the sand. Although a mature technology, there are still few instances of sandtables being used in surface analysis. Fundamentally there has not been any reported formal assessment of how well sandtables perform in an educational context compared to other conventional learning environments. We compared learning outcomes from using an AR sandtable versus a conventional 3D GIS to convey key concepts in terrain and hydrological analyses via usability and knowledge testing. Overall results from students at a research-intensive New Zealand university reveal a faster task performance and more learning satisfaction when using the sandtable to undertake experimental tasks. Effectiveness and knowledge quiz results revealed no significant difference between the technologies though there was a trend for more accurate answers with 3D GIS tasks. Student learning wise, the sandtable integrated core concepts (especially morphometry) more effectively though both technologies were otherwise similar. We conclude that sandtables have high potential in geospatial teaching, fostering accessible and engaging means of introducing terrain and hydrological concepts, prior to undertaking a more accurate and precise surface analysis with 3D GIS.

The mobility and ubiquity of mobile head-mounted displays make them a promising platform for telepresence research as they allow for spontaneous and remote use cases not possible with stationary hardware. In this work we present a system that provides immersive telepresence and remote collaboration on mobile and wearable devices by building a live spherical panoramic representation of a user's environment that can be viewed in real time by a remote user who can independently choose the viewing direction. The remote user can then interact with this environment as if they were actually there through intuitive gesture-based interaction. Each user can obtain independent views within this environment by rotating their device, and their current field of view is shared to allow for simple coordination of viewpoints. We present several different approaches to create this shared live environment and discuss their implementation details, individual challenges, and performance on modern mobile hardware; by doing so we provide key insights into the design and implementation of next generation mobile telepresence systems, guiding future research in this domain. The results of a preliminary user study confirm the ability of our system to induce the desired sense of presence in its users.

Determining relative camera pose is a fundamental problem in computer vision, and pose is often computed from feature correspondences. For point features, a minimum of five correspondences are required to determine the pose between two calibrated cameras, and eight corresponding points can be used to form a linear solution. However, most feature detectors used in practice produce points with an associated orientation. This work demonstrates that with oriented features the relative pose of two cameras can be computed from just four point correspondences, or seven with a linear solution. These new four- and seven-point algorithms do not require any additional sensors or parameters, but exploit information (feature orientation) that is already computed by most existing structure-from-motion systems. On the DTU multi-view stereo data set the four-point algorithm is shown to be 55% faster than the five-point algorithm, and the seven-point linear algorithm gives a 43% speed improvement over the eight-point algorithm.

The potential lack of fairness in the outputs of machine learning algorithms has recently gained attention both within the research community as well as in society more broadly. Surprisingly, there is no prior work developing tree-induction algorithms for building fair decision trees or fair random forests. These methods have widespread popularity as they are one of the few to be simultaneously interpretable, non-linear, and easy-to-use. In this paper we develop, to our knowledge, the first technique for the induction of fair decision trees.We show that our "Fair Forest" retains the benefits of the tree-based approach, while providing both greater accuracy and fairness than other alternatives, for both "group fairness'' and "individual fairness.'' We also introduce new measures for fairness which are able to handle multinomial and continues attributes as well as regression problems, as opposed to binary attributes and labels only. Finally, we demonstrate a new, more robust evaluation procedure for algorithms that considers the dataset in its entirety rather than only a specific protected attribute.

Sparse bundle adjustment (SBA) is a key but time- and memory-consuming step in three-dimensional (3D) reconstruction. In this paper, we propose a 3D point-based distributed SBA algorithm (DSBA) to improve the speed and scalability of SBA. The algorithm uses an asynchronously distributed sparse bundle adjustment (A-DSBA) to overlap data communication with equation computation. Compared with the synchronous DSBA mechanism (SDSBA), A-DSBA reduces the running time by 46%. The experimental results on several 3D reconstruction datasets reveal that our distributed algorithm running on eight nodes is up to five times faster than that of the stand-alone parallel SBA. Furthermore, the speedup of the proposed algorithm (running on eight nodes with 48 cores) is up to 41 times that of the serial SBA (running on a single node).

This letter introduces the LOOP binary descriptor (local optimal-oriented pattern) that encodes rotation invariance into the main formulation itself. This makes any post processing stage for rotation invariance redundant and improves on both accuracy and time complexity. We consider fine-grained lepidoptera (moth/butterfly) species recognition as the representative problem since it involves repetition of localized patterns and textures that may be exploited for discrimination. We evaluate the performance of LOOP against its predecessors as well as few other popular descriptors. Besides experiments on standard benchmarks, we also introduce a new small image dataset on NZ Lepidoptera. LOOP performs as well or better on all datasets evaluated compared to previous binary descriptors. The new dataset and demo code of the proposed method are available through the lead author's academic webpage and GitHub.

Approximate$k$Nearest Neighbours (A$k$NN) search is widely used in domains such as computer vision and machine learning. However, A$k$NN search in high-dimensional datasets does not scale well on multicore platforms, due to its large memory footprint. Parallel A$k$NN search using space subdivision for filtering helps reduce the memory footprint, but its loss of precision is unstable. In this paper, we propose a new data filtering method—PCAF—for parallel A$k$NN search based on principal component analysis. PCAF improves on previous methods, demonstrating sustained, high scalability for a wide range of high-dimensional datasets on both Intel and AMD multicore platforms. Moreover, PCAF maintains highly precise A$k$NN search results.

Sclera segmentation as an ocular biometric has been of an interest in a variety of security and medical applications. The current approaches mostly rely on handcrafted features which make the generalisation of the learnt hypothesis challenging encountering images taken from various angles, and in different visible light spectrums. Convolutional Neural Networks (CNNs) are capable of extracting the corresponding features automatically. Despite the fact that CNNs showed a remarkable performance in a variety of image semantic segmentations, the output can be noisy and less accurate particularly in object boundaries. To address this issue, we have used Conditional Random Fields (CRFs) to regulate the CNN outputs. The results of applying this technique to sclera segmentation dataset (SSERBC 2017) are comparable with the state of the art solutions.

Approximate k Nearest Neighbours (AkNN) search is widely used in domains such as computer vision and machine learning. However, AkNN search in high dimensional datasets does not work well on multicore platforms. It scales poorly due to its large memory footprint. Current parallel AkNN search using space subdivision for filtering helps reduce the memory footprint, but leads to loss of precision. We propose a new data filtering method -- PCAF -- for parallel AkNN search based on principal components analysis. PCAF improves on previous methods by demonstrating sustained, high scalability for a wide range of high dimensional datasets on both Intel and AMD multicore platforms. Moreover, PCAF maintains high precision in terms of the AkNN search results.

We present our efforts toward enabling a wearable sensor system that allows for the correlation of individual environmental exposures with physiologic and subsequent adverse health responses. This system will permit a better understanding of the impact of increased ozone levels and other pollutants on chronic asthma conditions. We discuss the inefficiency of existing commercial off-the-shelf components to achieve continuous monitoring and our system-level and nano-enabled efforts toward improving the wearability and power consumption. Our system consists of a wristband, a chest patch, and a handheld spirometer. We describe our preliminary efforts to achieve a submilliwatt system ultimately powered by the energy harvested from thermal radiation and motion of the body with the primary contributions being an ultralow-power ozone sensor, an volatile organic compounds sensor, spirometer, and the integration of these and other sensors in a multimodal sensing platform. The measured environmental parameters include ambient ozone concentration, temperature, and relative humidity. Our array of sensors also assesses heart rate via photoplethysmography and electrocardiography, respiratory rate via photoplethysmography, skin impedance, three-axis acceleration, wheezing via a microphone, and expiratory airflow. The sensors on the wristband, chest patch, and spirometer consume 0.83, 0.96, and 0.01 mW, respectively. The data from each sensor are continually streamed to a peripheral data aggregation device and are subsequently transferred to a dedicated server for cloud storage. Future work includes reducing the power consumption of the system-on-chip including radio to reduce the entirety of each described system in the submilliwatt range.

Determining relative camera pose is a fundamental task in many computer vision systems. Various algorithms have been proposed for determining relative camera pose from feature correspondences, usually based on point correspondences. These correspondences are commonly found using SIFT and similar feature detectors, and Random Sample and Consensus (RANSAC) is commonly applied to identify and remove incorrect matches. The 'points' found by these feature detectors, however, are not simply two-dimensional image locations -- they have scale and orientation as well. The orientation associated with SIFT-like features can be used to quickly identify many incorrect pose hypotheses in a RANSAC process. This can reduce the number of camera poses that are evaluated with more expensive techniques, with a corresponding decrease in pose estimation time.

k Nearest Neighbours (k-NN) search is a fundamental problem in many computer vision and machine learning tasks. These tasks frequently involve a large number of high-dimensional vectors, which require intensive computations. Recent research work has shown that the Graphics Processing Unit (GPU) is a promising platform for solving k-NN search. However, these search algorithms often meet a serious bottleneck on GPUs due to a selection procedure, called k-selection, which is the final stage of k-NN and significantly affects the overall performance. In this paper, we propose new data structures and optimization techniques to accelerate k-selection on GPUs. Three key techniques are proposed: Merge Queue, Buffered Search and Hierarchical Partition. Compared with previous works, the proposed techniques can significantly improve the computing efficiency of k-selection on GPUs. Experimental results show that our techniques can achieve an up to 4:2× performance improvement over the state-of-the-art methods.

k Nearest Neighbors (k-NN) search is a widely used category of algorithms with applications in domains such as computer vision and machine learning. Despite the desire to process increasing amounts of high-dimensional data within these domains, k-NN algorithms scale poorly on multicore systems because they hit a memory wall. In this paper, we propose a novel data filtering strategy for k-NN search algorithms on multicore platforms. By excluding unlikely features during the k-NN search process, this strategy can reduce the amount of computation required as well as the memory footprint. It is complementary to the data selection strategies used in other state-of-the-art k-NN algorithms. A Subspace Clustering for Filtering (SCF) method is proposed to implement the data filtering strategy. Experimental results on four k-NN algorithms show that SCF can significantly improve their performance on three modern multicore platforms with only a small loss of search precision.

Stereo reconstruction is challenging in scenes with many similar-looking objects, as matches between features are often ambiguous. Features matched incorrectly lead to an incorrect 3D reconstruction, whereas if correct matches are missed, the reconstruction will be incomplete. Previous systems for selecting a correspondence (set of matched features) select either a maximum likelihood correspondence, which may contain many incorrect matches, or use some heuristic for discarding ambiguous matches. In this paper we propose a new method for selecting a correspondence: we select the correspondence which minimises an expected loss function. Match probabilities are computed by Gibbs sampling, then the minimum expected loss correspondence is selected based on these probabilities. A parameter of the loss function controls the trade off between selecting incorrect matches versus missing correct matches. The proposed correspondence selection method is evaluated in a model-based framework for reconstructing branching plants, and on simulated data. In both cases it outperforms alternative approaches in terms of precision and recall, giving more complete and accurate 3D models.

K Nearest Neighbors (k-NN) search is a widely used category of algorithms with applications in domains such as computer vision and machine learning. With the rapidly increasing amount of data available, and their high dimensionality, k-NN algorithms scale poorly on multicore systems because they hit a memory wall. In this paper, we propose a novel data filtering strategy, named Subspace Clustering for Filtering (SCF), for k-NN search algorithms on multicore platforms. By excluding unlikely features in k-NN search, this strategy can reduce memory footprint as well as computation. Experimental results on four k-NN algorithms show that SCF can improve their performance on two modern multicore platforms with insignificant loss of search precision.

This paper proposes a novel approach to explore emergent patterns in images in an unsupervised setting. We consider emergent patterns to be sets of co-occurring visual words that appear together more often than chance would indicate. Rather than focusing on finding ways to learn a large number of objects or their categories we focus on analyzing behavior associated with emergent patterns. We show that these patterns emerge from the data and in some cases relate to object identifiers. We extract SIFT descriptors [1] and then cluster them to represent each image as a bag-of-words. To encode co-occurrences between visual words we represent them as edges of a graph which are weighted by the number of images containing a particular co-occurrence. Performing a statistical analysis on weights of the edges identifies words which co-occur significantly more often than expected. These highly co-occurring nodes produce clusters in the graph which can be separated using normalized cuts. Applying normalized cuts reveals that in simple images datasets these emergent clusters can identify object classes. Results on more complex datasets like Caltech101 [2] show that interesting patterns other than object classes can also emerge from the data.

3D reconstruction from multiple cameras is challenging in some environments because of ambiguous matches between similar-looking features. These ambiguities can be resolved by projecting a structured light pattern into the scene, and detecting points in the light pattern in each image. Robust detection of the structured light pattern is hard because of variations in object colour and lighting within the scene, however for specific applications, training data can easily be collected and labelled, enabling the detection problem to be solved using machine learning techniques. We demonstrate the application of a Support Vector Machine (SVM) to detect laser light patterns projected into images of vines, using Feature Subset Selection to design a feature descriptor. A descriptor is computed for every candidate pixel, and the SVM determines if each descriptor is part of the laser line pattern. On test images, the proposed detector achieves 99.4% precision at 90% recall, outperforming a detector which uses only one pixel's colour.

Despite recent attention paid to feature detection and description methods, the basic criterion for feature matching has remained largely unchanged. Current techniques typically rely on the feature description vector extracted by SIFT or similar descriptors. Many feature detectors, however, also estimate the orientation and scale of features, which are valuable guides to correspondence. This paper outlines a technique for exploiting relative orientation and scale to improve feature matching performance. It is shown that these cues can significantly improve matching performance, as measured by the percentage of inliers, across a range of different image transforms and object recognition rates.

Parallel programming is the mainstream for today's HPC applications. Programmers need to parallelize their programs to achieve better performance on multicore systems. However, due to a lack of good understanding of parallelism in algorithms, scheduling policy in runtime systems, and multicore architectures, programmers usually find it very hard to write high-performance, scalable programs on these parallel platforms. Although using a parallelized library written by experts can reduce the amount of work for coding, it does not automatically guarantee good performance according to our study. A better understanding of parallelism in algorithms, the OS/runtime systems, and hardware architectures is necessary if programmers wish to further improve performance. In this paper, we use SIFT-based feature matching within large-scale image collections to show the importance of three factors-the level of parallelism, scheduling policy, and memory architecture-that affect the performance of large-scale feature matching on multicore systems. We demonstrate experimental results using programs based on OpenCV and OpenMP, which are executed on both 16-core and 64-core machines. From our experimental results, we find that images with a large number of features achieve poor scalability on the 64-core machine due to a poor cache utilization. To address this issue of cache performance, we propose a Divide-and-Merge algorithm that divides the feature space into several small sub-spaces so that they fit within the cache. Our experiments show that the performance tuning addressing all of the three factors improves the speedup of feature matching from 10.6× to 21.5× on the 64-core machine. While the speedup is improved by 103%, the scalability of the feature matching algorithm is improved by up to 6.45 times on the 64-core machine with our performance tuning. Our study indicates that performance tuning on multicore systems is very challenging even for a simple image processing algorithm.

This paper presents an Image Based Visual Servo control design for Fixed Wing Unmanned Aerial Vehicles tracking locally linear infrastructure in the presence of wind using a body fixed imaging sensor. Visual servoing offers improved data collection by posing the tracking task as one of controlling a feature as viewed by the inspection sensor, although is complicated by the introduction of wind as aircraft heading and course angle no longer align. In this work it is shown that the effects of wind alter the desired line angle required for continuous tracking to equal the wind correction angle as would be calculated to set a desired course. A control solution is then sort by linearizing the interaction matrix about the new feature pose such that kinematics of the feature can be augmented with the lateral dynamics of the aircraft, from which a state feedback control design is developed. Simulation results are presented comparing no compensation, integral control and the proposed controller using the wind correction angle, followed by an assessment of response to atmospheric disturbances in the form of turbulence and wind gusts.

This paper proposes a novel solution to the problem of scale drift in single-camera simultaneous localization and mapping, based on recognizing and measuring objects. When reconstructing the trajectory of a camera moving in an unknown environment, the scale of the environment, and equivalently the speed of the camera, is obtained by accumulating relative scale estimates over sequences of frames. This leads to scale drift: errors in scale accumulate over time. The proposed solution is to learn the classes of objects that appear throughout the environment and to use measurements of the size of these objects to improve the scale estimate. A bag-of-words-based scheme to learn object classes, to recognize object instances, and to use these observations to correct scale drift is described and is demonstrated reducing accumulated errors by 64% while navigating for 2.5 km through a dynamic outdoor environment.

Visual exposure modelling establishes the extent to which a nominated feature may be seen from a specified location. The advent of high-resolution light detection and ranging (LiDAR)-sourced elevation models has enabled visual exposure modelling to be applied in urban regions, for example, to calculate the field of view occupied by a landmark building when observed from a nearby street. Currently, visual exposure models access a single surface elevation model to establish the lines of sight (LoSs) between the observer and the landmark feature. This is a cause for concern in vegetated areas where trees are represented as solid protrusions in the surface model totally blocking the LoSs. Additionally, the observer's elevation, as read from the surface model, would be incorrectly set to the tree top height in those regions. The research presented here overcomes these issues by introducing a new visual exposure model, which accesses a bare earth terrain model, to establish the observer's true elevation even when passing through vegetated regions, a surface model for the city profile and an additional vegetation map. Where there is a difference between terrain and surface elevations, the vegetation map is consulted. In vegetated areas the LoS is permitted to continue its journey, either passing under the canopy with clear views or partially through it depending on foliage density, otherwise the LoS is terminated. This approach enables landmark visual exposure to be modelled more realistically, with consideration given to urban trees. The model's improvements are demonstrated through a number of real-world trials and compared to current visual exposure methods.

This paper presents an evaluation of airborne sensors for use in vegetation management in power-line corridors. Three integral stages in the management process are addressed, including the detection of trees, relative positioning with respect to the nearest power line, and vegetation height estimation. Image data, including multispectral and high resolution, are analyzed along with LiDAR data captured from fixed-wing aircraft. Ground truth data are then used to establish the accuracy and reliability of each sensor, thus providing a quantitative comparison of sensor options. Tree detection was achieved through crown delineation using a pulse-coupled neural network and morphologic reconstruction applied to multispectral imagery. Through testing, it was shown to achieve a detection rate of 96%, while the accuracy in segmenting groups of trees and single trees correctly was shown to be 75%. Relative positioning using LiDAR achieved root-mean-square-error (rmse) values of 1.4 and 2.1 m for cross-track distance and along-track position, respectively, while direct georeferencing achieved rmse of 3.1 m in both instances. The estimation of pole and tree heights measured with LiDAR had rmse values of 0.4 and 0.9 m, respectively, while stereo matching achieved 1.5 and 2.9 m. Overall, a small number of poles were missed with detection rates of 98% and 95% for LiDAR and stereo matching.

RANSAC (Random Sample Consensus) [2] is a popular algorithm in computer vision for fitting a model to data points contaminated with many gross outliers. Traditionally many small hypothesis sets are chosen randomly; these are used to generate models and the model consistent with most data points is selected. Instead we propose that each hypothesis set chosen is the one most likely to be correct, conditional on the knowledge of those that have failed to lead to a good model. We present two algorithms, BaySAC and SimSAC, to choose this most likely hypothesis set. A common use for RANSAC is for estimating the essential matrix describing the relative position of two cameras, which can be computed from stereo matches between two images. Given a set of these stereo correspondences the classic RANSAC sampling algorithm selects random subsets of five points, and for each subset computes all possible essential matrices. Each essential matrix is checked against all correspondences until one compatible with a large number of correspondences is found. RANSAC for essential matrix estimation can be a costly part of realtime Visual Navigation schemes because of the large number of hypothesis sets that must be tried before finding one uncontaminated by outliers. This number may be reduced considerably if outlier probabilities can be estimated, e.g. from stereo correspondence match strengths. PROSAC (Progressive Sample Consensus) [1] ranks data points by prior probability then selects subsets in (approximate) order of prior likelihood. Alternatively Guided-MLESAC [3] selects random subsets where each data-point is selected with probability in proportion to its prior inlier likelihood. These sampling methods fail to take into account is the information gained by testing hypothesis sets and finding them to be contaminated by outliers, unlike the two methods proposed here which are based on the following observation: a hypothesis set leading to a model consistent with few data points probably contains one or more outliers (the alternative possibility is that it contains a degenerate configuration of inliers). Hypothesis sets with one or more data points in common with this set are also now less likely, as they are likely to include the same outlier(s). Ideally at each time we will choose one of the hypothesis set that is most likely to contain no outliers based on the prior probabilities and the history of contaminated samples. This strategy minimises the number of hypotheses that must be tested before finding one consisting entirely of inliers. Unfortunately a closed-form solution for this posterior probability is algebraically intractable. Instead we present two methods of approximating this probability, both of which are shown to work well in practice.

Particle filtering provides a well-developed and widely adopted approach to visual tracking. For effective tracking in real-world environments the particle set must sample widely enough that it can represent alternative target states in areas of ambiguity. It must not, however, become diffuse, spreading across the image plane rather than clustering around the object(s) of interest. A key issue in the design of particle filter-based trackers is how to manage the spread of the particle set to balance these conflicting requirements. To be computationally efficient, balance must be achieved with as small a particle set as reasonably possible. A number of hybrid particle filter/mean-shift trackers have recently been proposed. We believe that their strength lies in their ability to alternately disperse and cluster particles together, providing both a degree of balance and a reduced particle set. We present a novel hybrid of the annealed particle filter and kernel mean-shift algorithms that emphasises this behaviour. The algorithm has been applied to a wide variety of artificial and real image sequences. The method has performance and efficiency advantages over both pure kernel mean-shift and particle filtering trackers and existing hybrid algorithms

This paper describes and evaluates a novel set of approaches to handle situations where multiple distinct and visually differing objects are tracked, such as tracking of people and objects they are manipulating. Unlike tracking of multiple similar objects, visually different interacting objects can provide an opportunity to improve the tracking accuracy. These approaches are designed for use with Condensation/Particle Filter based algorithms, and allow drop-in replacement of tracker modules for each object type tracked. They use information about the relationships and interactions between objects to improve the tracking, rather than in order to distinguish between the objects, as in current algorithms. They are also designed to be highly efficient, for real time use. The approaches are tested on a challenging set of real data and achieve tracking performance similar to using a single very high dimensional tracker, but with vastly reduced complexity and hence much better time performance. 1

A combined tracking method using the Kalman filter and Hough transform is presented. An extended Kalman filter is used to model the parameters and motion of a set of lines detected in a Hough space The integration of these two techniques gives a number of advantages. The use of a Hough transform provides resilience to noise and partial occlusion, and the Kalman filter’s ability to predict future states is used to reduce the computational load of line detection. Analysis of the tracker from synthetic data shows that it is robust to noise, occlusion, and deviations from the constant motion model underlying the Kalman filter. Tracking results from video sequences illustrate its applicability to real-world domains. 1

Long image sequences provide a wealth of information, which means that a compact representation is needed to efficiently process them. In this paper a novel representation for motion segmentation in long image sequences is presented. This representation – the feature interval graph – measures the pairwise rigidity of features in the scene. The feature interval graph is re-cursively computed, making it a compact representation, and uses an interval model of uncertainty. The feature interval graph forms the basis for new al-gorithms for motion segmentation and occlusion analysis. Results of these algorithms are presented on synthetic and laboratory scenes. 1

Evaluating the performance of optical flow algorithms has been difficult because of the lack of ground-truth data sets for complex scenes. We describe a simple modification to a ray tracer that allows us to generate ground-truth motion fields for scenes of arbitrary complexity. The resulting flow maps are used to assist in the comparison of eight optical flow algorithms using three complex, synthetic scenes. Our study found that a modified version of Lucas and Kanade's algorithm has superior performance but produces sparse flow maps. Proesmans et al.'s algorithm performs slightly worse, on average, but produces a very dense depth map. 1 Introduction Seventeen years have passed since Horn and Schunck published their influential paper on the calculation of optical flow [4]. Since then, a substantial amount of research has been devoted to finding ways to calculate optical flow more efficiently and more accurately. Optical flow extraction has been proposed as a preprocessing step...