The harmony search algorithm was developed in 2001 as a heuristic optimisation algorithm for use in diverse optimisation problems. After its introduction it was extensively used in multiple engineering disciplines with great success. In order to demonstrate the value of harmony search in computer vision applications I developed four novel algorithms based on harmony search that efficiently solves three problems that are commonly found in computer vision, namely visual tracking, visual correspondence matching and binary image restoration. Computer vision is a large discipline that includes solving many different kinds of optimisation problems. Many of these optimisation problems are discontinuous with derivative information difficult or impossible to come by. The most common solution is to use population based statistical optimisation algorithms like the particle filter, genetic algorithms, PSO, etc. but harmony search has never been investigated as a possible alternative. This is surprising since harmony search has been shown to be superior to these methods in several other engineering disciplines. I therefore aim to show that harmony search deserves to be included in the computer vision researcher's toolbox of optimisation algorithms through the introduction of four novel algorithms based on harmony search that solve three diverse problems in computer vision. First the harmony filter (HF) is introduced as a visual tracking algorithm that is shown to be superior to the particle filter and the unscented Kalman filter (UKF) in both speed and accuracy for robust tracking in challenging situations. The directed correspondence search (DCS) algorithm is then introduced as a solution to the visual correspondence problem. Finally, two algorithms, counterpoint harmony search (CHS) and largest error first harmony search (LEFHS), are introduced for the blind deconvolution of binary images. Comparative results from these algorithms are very promising. The harmony filter was compared with the particle filter and the UKF both of which have been extensively used in visual tracking. In challenging situations consisting of rapid and erratic target movement, extended periods of total and partial occlusion and changing light conditions, the HF proved to be more accurate and faster on average than both the particle filter and the UKF. Under various conditions I show that the HF is at least 2 times faster than a UKF implementation and 4 times faster than a particle filter implementation (using 300 particles). While there are fewer algorithms specialising in the blind deconvolution of binary images, CHS and LEFHS were compared with a current state-of-the-art method and proved to be more robust to noise and more accurate. LEFHS is the only algorithm currently available that can recover a 24 x 12 binary image using blind deconvolution to 100% accuracy without putting constraints on the point spread function (blurring kernel). During the development of these algorithms several valuable insights into the inner workings of harmony search were discovered. In each application harmony search had to be adapted in a different way and with each new adaptation a deeper understanding of the advantages of harmony search is revealed. Knowing which components may be modified without degrading performance is key to adapting harmony search for use in diverse problems and allows one to use harmony search in situations it was not originally designed for without losing its superior performance. These insights and the adaptation strategies that they lead to are the main contribution of this thesis.

No description available.

No description available.

Localization and mapping are important abilities for any robot to have if it wants to navigate intelligently in the real world. The goal of the research designed in this thesis was to develop a practical embedded visual odometer that utilized common features found in real world environments. The visual odometer is a system that measures the self motion of a mobile robot using visual feeback. The developed visual odometer was tested on a custom mobile robot in several different tests that were derived from the robotic soccer domain. This system’s performance was compared to two other systems. These systems were a KLT feature tracker based robot and a commercial shaft encoder based robot. The results of the completed tests showed that the developed visual odometer’s performance was less than expected. It also showed that this system has good potential. As well, the test results showed the limitations of using a KLT feature tracker based robot and that the commercial shaft encoder based robot also had performance less than expected.

NEPTUNE Canada possesses a large collection of video data for monitoring marine life. Such data is important for marine biologists who can observe species in their natural habitat on a 24/7 basis. It is counterproductive for researchers to manually search for the events of interest (EOI) in a large database. Our study aims to perform the automatic detection of the EOI de ned as animal motion. The output of this approach is in a summary video clip of the original video fi le that contains all salient events with their associated start and end frames. Our work is based on Laptev [1] spatio-temporal interest points detection method. Interest points in the 3D spatio-temporal domain (x,y,t) require frame values in local spatio-temporal volumes to have large variations along all three dimensions. These local intensity variations are captured in the magnitude of the spatio-temporal derivatives. We report experimental results on video summarization using a database of videos from Neptune Canada. The eff ect of several parameters on the performance of the proposed approach is studied in detail. / Graduate

One of the fundamental problems in computer vision is the identification of objects from their shape. The research reported in this thesis is directed toward the development of a scheme for representing the shape of an object which allows it to be recognised both quickly and robustly across a wide range of viewing conditions. Given a shape described by a set of primitive elements, eg. straight line segments, the proposed scheme involves using a histogram to record the distribution of geometric features, eg. angle and distance, measured between pairs of primitives. This form of shape representation has a number advantages over previously proposed schemes. Foremost among these is the fact that it is able to produce local representations of shape, based on individual line segments. Recognition based on such representation is robust to the problems arising in cluttered scenes. Representations produced by the scheme are also invariant to certain object transformations, they degrade gracefully as the shape is fragmented and are strong enough to support discrimination between dissimilar objects. By treating the histogram recording a geometric feature distribution as a feature vector it is possible to match shapes using techniques from statistical pattern classification. This has the advantage that optimal matching accuracy can be achieved using processing which is both simple and uniform. The approach is therefore ideally suited to implementation in dedicated hardware. A detailed analysis is undertaken of the effect on recognition of changes in the description of a shape caused by fragmentation noise, scene clutter and sensor error. It is found that the properties of both the representation and matching components of the system combine to ensure that recognition is, in theory, unaffected by fragmentation noise, while it is maintained to very high levels of scene clutter. The factors which determine the effect of sensor error on the performance of the recognition system are fully analysed. The ability of the representational scheme to support object recognition is demonstrated in a number of different domains. The recognition of both 2D and 3D objects from a fixed viewpoint is demonstrated in conditions of severe fragmentation noise, occlusion and clutter. The scheme is then shown to extend straightforwardly to the representation of 3D shape. This is exploited to perform recognition and localisation of 3D objects from an arbitrary viewpoint, based on the matching of 3D scene and ,model shape descriptions. Finally, the use of the scheme within a multiple view-based approach to 3D object recognition is demonstrated.

No description available.

No description available.

No description available.

No description available.