الفهرس | Only 14 pages are availabe for public view |
Abstract isual object tracking remains a challenge facing an intelligent control system. Variety of applications serves many purposes such as surveillance and security. The developed technology faces plenty of obstacles that should be addressed including occlusion, fast motion, motion blur, background clutter, non-rigid object deformation, low resolution, illumination variation, in-plane rotation, and scale variation. In visual tracking, online learning techniques are most common due to their efficiency for most video sequences. So, it is essential to make a fast and efficient algorithm with minimum error. Many object tracking techniques have emerged. However, the drifting problem when noisy updates appear has been a stumbling block for the plurality of relevant mechanisms. But this problem can now be overcome by updating the classifiers This thesis aims introduces a new hybrid model for object tracking system for partially or fully occluded objects. The suggested system is called the occluded objects tracking system (OOTS). It is a hybrid system between two algorithms: a fast technique Circulant Structure Kernels with Color Names (CSK-CN) and the efficient algorithm occlusion-aware Real-time Object Tracking (ROT). Another model will be introduced to improve the previous system that hybrid OOTS with a CNN-based tracking system called adaptive structural convolutional network Abstract (ASCT). The suggested system is evaluated with a standard visual tracking benchmark dataset (OTB-2015). The experimental results proved that the suggested system is more trustworthy, and gives efficient tracking results than other methods. The first Proposed model reduced the average of center location error (CLE) from 62.7 to 57.8 and improved the real-time performance from 12.07 frame per second (FPS) to 47.6. The second Proposed model reduced CLE from 57.8 to 42.5, and it still achieves real-time performance. |