الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 135 |  |  |
| | | from | 135 | | |
Abstract
Multi-object tracking (MOT) is a critical research area in computer vision. A very mature effort has been introduced recently in this area. However, occlusion is still a challenge. As occlusion always happens in real life, solving such a challenge would be useful for the researcher community. The human brain can effortlessly perform vision processes using the visual system, which helps solve MOT problems. Few algorithms simulate human strategy in solving MOT problems. Therefore, devising a method that simulates human activity in vision has become a good choice for improving MOT results. The main objective of this thesis is to introduce a brain-based strategies algorithm (Merging Semantic Attributes and Appearance Feature MSA-AF) to solve the occlusion in MOT. Eight brain strategies have been studied from a cognitive perspective. Then, imitated in the algorithm. Two of these strategies gave the algorithm novelty and outstanding results, rescue saccades and stimulus attributes. First, rescue saccades are imitated by detecting the occlusion state in each frame, representing the critical situation that the human brain saccades toward. Then, stimulus attributes are mimicked by using semantic attributes to reidentify the person in these occlusion states. MSA-AF algorithm favourably performs on the MOT17 dataset. In addition, this thesis is releasing a new dataset of 40,000 images, 190,000 annotations and four classes that are used to train the detection model to detect occlusion and semantic attributes. PGA dataset is unique regarding classes, meaning no other dataset has those classes, and images, which will benefit researchers on MOT, detection, and fashion-based detection. Experimental results demonstrate that the PGA dataset achieves outstanding performance on the Scaled YOLOv4 detection model by achieving 0.89 mAP 0.5. The quantitative performance of MSA-AF has been measured and compared to five from the most recent state-of-the-art trackers. These results show the MSA-AF algorithm’s impact on creating a stable tracker by decreasing the Mostly Lost (ML) metric and increasing the MOT Precision (MOTP) at the same time.