الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 82 |  |  |
| | | from | 82 | | |
Abstract
Video tracking is one of the most active research topics recently that has many applications such as teleconferencing, surveillance, and security. The research proposes two algorithms: SURF- Particle Algorithm and Adaptive Variable-Size Search Window Algorithm. SURF- Particle tracker uses the discriminative interest points generated by the SURF descriptor as the initial particles/ samples to be fed into particle filter instead of choosing these particles randomly as done in traditional simple particle filter. The proposed SURF-Particle tracker proved to be more efficient, reliable and accurate than traditional particle filter and SIFT-Particle tracker. The proposed algorithm solved the sample degeneration problem of particle filter as well.
Adaptive variable-size search window algorithm rose from the challenge in video tracking of exactly determining the location of the tracked object within each frame. Usually, a search window is to be proposed around the tracked object. The tracking algorithm should search for this search window in every frame of the video. Most of tracking algorithms make use of a fixed size search window regardless of the tracked object scale change over time. The fact is that too small search window may lose details of the tracked object. On the other hand, undue increase of computational complexity is resulted of inaccurate large search window. Even if the tracked object is partially or completely occluded the proposed algorithm should locate the expected location of it in an efficient way. The proposed search window updating algorithm is based on speeded up robust features (SURF). It makes use of the position information of the extracted SURF points to update the size and location of the search window in the following frame and so forth. The proposed algorithm produces a search window that is more fitted to the tracked object than search windows produced by common tracking algorithms such as mean shift do. Less computational time in the search window is an added value. Prediction of the exact location of the tracked object under occlusion is more precise than existing algorithms. At the end of our research, we integrated the two proposed algorithms into one video tracking system to measure the overall accuracy. The system proved to work with high performance in most of the cases as shown in the results.