| Dr Steven Mills | ||||
| Home | Teaching | Research | Publications | |
Publications
AbstractsFinding Fencing Foils using the Hough TransformWe present a method for the analysis of images of foil fencing. This is a challenging domain because of the speed of motion, which leads to significant motion blur. Our method detects the blade in images of a fencer, and is based on the Hough transform. In experiments under controlled conditions it is found to have a 94% success rate on two unseen sequences of 500 frames. [back to top] Switching Tempalte Fitting Methods during Articulated Object TrackingThis paper describes two methods of fitting deformable templates when tracking articulated objects using particle filters. One method fits a template to each of the links of an articulated object in a hierarchical way. The method first fits a template for the base of the articulated object and then fits a template for each of the links deeper in the hierarchy. The second method fits the whole articulated object as a rigid object, and then refines the fitting for each of the links of the articulated object in a hierarchical way, starting from the base. Advantages and disadvantages of each method are discussed and a way of combining the best of each method in a single tracker is presented. [back to top] Tracking in a Hough Space with the Extended Kalman FilterA 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. [back to top] Texture Unmapping - Combining Parametric and Non-Parametric Techniques for Image ReconstructionWe present a new technique, which we call "texture unmapping", and its application to image reconstruction. A parametric model of an image is made, and then subtracted from the image to leave a non-parametric component. The two components may then be reconstructed independently and then recombined to give a final reconstruction. Results of the new technique are given on a range of images. [back to top] Object Tracking Through a Hough SpaceA visual tracking method is introduced which performs no direct tracking of any feature in the image plane. Instead, images are transformed using a variant of the Hough technique and features are tracked by the parameters which describe them. This enables features to be combined and constrained by their structure, allowing tracking of more complex shapes with a controlled degree of flexibility and distortion. Results are presented tracking rectangular structures in a variety of image sequences. [back to top] Motion Tracking in Long Image SequencesLong 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 recursively computed, making it a compact representation, and uses an interval model of uncertainty. The feature interval graph forms the basis for new algorithms for motion segmentation and occlusion analysis. Results of these algorithms are presented on synthetic and laboratory scenes. [back to top] Motion Segmentation in Long Image SequencesThis thesis presents new methods fo the analysis of long image sequences. Long sequences contain large amounts of data, making memory a concern, but allow decisions to be delayed until more observations are made. The methods presented are based on a graph structure, the feature interval graph, which is constructed from information provided by low-level processes such as stereo and motion tracking. The graph's vertices represent features in the scene, and the edges represent hypothesised rigid relationships between them. The distance between features is monitored over time, and if the distance between two features changes then the corresponding edge is removed from the graph. In this way the graph evolves over time to represent the rigid structure of the scene. Uncertainty in the measurements made at a low level is represented using interval quantities. This uncertainty is propagated through subsequent computations, to the feature interval graph. Intervals also provide a mechanism for combining uncertain measurements over time. The use of intervals provides an alternative to statistical approaches, such as the Kalman filter, which have been previously applied to image sequence analysis. The feature interval graph forms the basis for a novel motion segmentation algorithm. This segmentation method is based on finding small rigid substructures in the graph, and is robust to a small number of errors in the graph. These errors arise when incorrect or insufficient information is provided by low-level processes. A new approach to reasoning about partial occlusion is also proposed. This uses information from the motion segmentation to predict the locations of occluded points in the scene. These algorithms are presented on a variety of artificial and laboratory scenes. [back to top] Graph-Based Object HypothesisReasoning about three-dimensional scenes involves the analysis of large amounts of complex, time-varying information. This paper presents a novel representation for storing such information that provides a unified, intuitive framework for reasoning about rigid objects. The core structure is a graph that represents features of the scene as vertices, and stores distances between these features at the edges. These relationships evolve over time, as more information is presented to the system, and an approach to dealing with uncertain information using interval arithmetic is proposed. An algorithm for rigid object extraction is presented, based on this graph. The graph structure can also be used in support of some low-level tasks, and the problem of tracking three-dimensional features is considered. Results are presented for both artificial and real scenes. [back to top] Interval Computations in Computer VisionInterval arithmetic is a method for performing computations on measurements that are only known to within a fixed error range. As the measurements are combined mathematically, their error intervals are also combined, in a conservative fashion. While the technique of interval analysis has found uses in many areas of computing in recent years, it has not yet been applied heavily in computer vision. We describe the technique of interval arithmetic and show how its use can lead to a greater understanding of depth from stereo estimates. [back to top] Stereo-Motion Analysis of Image SequencesStereopsis and motion analysis have traditionally been treated as independent modules contributing to mid- or high-level visual processes. Recently there has been some suggestion that these two processes may interact at a very early stage. In this paper the possibility of combining of stereo vision and motion analysis is examined. Both of these processes are correspondence based and the combination of the two offers new constraints to these correspondences. The aim is to calculate the three-dimensional motion of points in a scene and to group points having similar motions into "objects". A graph-based grouping scheme is proposed and some of the problems associated with such an approach examined. The motion of these objects can then be estimated and used to predict where they will lie in the future. [back to top] | ||||