Here is the abstract from the PhD thesis...
Current multimedia research is creating systems that are capable of manipulating vast amounts of information. The data can represent text, sound, graphics, still images and image sequences (video). There are few tools for manipulating image data directly and the incorporation of images into large systems is awkward and time consuming.This thesis investigates methods for building motion analysis tools for use with multimedia systems. An object tracking tool is described that can track a user selected object through an arbitrary image sequence. The trajectory information can be used by the developer of a multimedia system to automatically place hypermedia buttons over the object as it moves. A tool to detect moving objects is also described, which is useful in applications where motion is being studied.
The object tracker uses a novel combination of Kalman filter based motion modelling and a matching method to locate the object in each frame. Several matching methods are compared, including template matching based on intensity, edge and corner features and a generalised Hough transform method that can track changes in object orientation. A new, shape independent histogram matching method is described that can track objects as they change size and shape. In addition, three motion modelling methods for enhancing the tracking performance are described and compared.
A library of data structures and an implementation methodology for building general purpose image analysis software are presented. The emphasis is on flexibility and portable code, allowing the system to be applied to many areas of image analysis. The library has been used to implement all the methods described in this thesis and is currently providing a basis for several on-going projects.