الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 109 |  |  |
| | | from | 109 | | |
Abstract
This thesis covers two topics in computer vision: object tracking and object Detection. Algorithms and methods were contributed for fast object tracking and object tracking in compressed videos. Object tracking is one of the key problems in visual surveillance systems as it provides the reliable motion behavior of the object of interest. Object tracking is also commonly used in content-based video retrieval where the video is indexed according to object motion information.The contribution presented an algorithm referred to as the hybrid iterated Kalman particle filter (HIKPF). The proposed algorithm is developed from unscented Kalman filter (UKF) and iterated extended Kalman filter (IEKF) to generate the proposal distribution leading to efficient use of the latest observations and generates more close approximation of the posterior probability density. Numerical simulation and experiment results show that HIKPF algorithmgives better accuracy results than the previously proposed algorithms such as (PF, PF-EKF, PF-UPF and PF-IEKF). It is also showed that the RMSE is the smallest than the other algorithms,which makes it a better choice for theproposal distribution.
Also, the second contribution is intended for the efficient object tracking to handle the problems of partial occlusion and illumination change.The proposed tracking algorithm is presented, which combines particle filter and optimized likelihood. Compared with the popular particle filters algorithm, the proposed algorithm gives better performance and tracking accuracy.The experimental results demonstrate that the proposed algorithm can effectively overcomes the problems of object occlusion and can track the color target efficiently in presence of illumination changes.
Finally, one of the critical tasks in object tracking is the tracking of fast-moving objects in complex environments, which contain cluttered background and scale change. So, a nother tracking algorithm is presented by using the joint color texture histogram to represent a target and then applying it to particle filter algorithm. The texture features of the object are extracted by using the local binary pattern (LBP) technique to represent the object. The proposed algorithm extracts effectively the edge and corner features in the target region, which characterize better and represent more robustly the target. As a result, this tracker is capable of tracking scale change and velocity change of the object.Also, the experiments showed that this new proposed algorithm produces excellent tracking results and outperforms other tracking algorithms.
We experiment with a number of videos and the results indicate that these algorithms work well under many complex and realistic situations. For the general treatment of a tracking system, object tracking and object segmentation are combined together to achieve a fully automated process. The performance of each component can affect the other.
The ultimate goal of object tracking and segmentation research is the creation of a surveillance system which operates robustly under various conditions. In the real world, there are many difficulties which prevent a system from maintaining high accuracy under complex conditions. The difficulties include: shadows, real-time requirements, occlusion, object appearance change, illumination change, etc. Algorithms developed in this thesis offer solutions to some of the difficulties we mentioned above such as tracking under illumination change, partial occlusion. So, the work presented is themain step in computer vision and surveillance system research towards the larger goal of building intelligent machines that can see the outside world and respond and communicate with people