Object tracking using surveillance camera

• Main Author: She, Hanqing
• Other Authors: Lap-Pui Chau
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/141724
• Institution: Nanyang Technological University

In recent years, due to the wide demand for multi-object tracking applications, many effective multi-object tracking methods have been produced. The whole process of multi-object tracking can be divided into two parts: one is object detection and the other is object tracking. And the evaluations of existing multi-object tracking methods usually separate the performance of object tracking and object detection. The same detection results by the same detectors are usually provided in order to compare the performance of the tracker fairly. Beyond that, in some applications, such as vehicle detection and tracking, the tracking process needs to be done in real-time. This gives us a challenge that the speed of our detection and tracking method should run as fast as the video frame. Multiple object tracking is usually done by tracking-by-detection approach. This requires detection to be done for each frame, before tracking-by-detection associates detections across frames to form tracks. The speed could be slow because current detection methods are based on CNN. One solution to this problem is that the detection algorithm does not run in every frame but runs after some frames interval to make sure that the speed of tracking algorithms can catch up with the frame rates, which uses the visual tracker. This thesis does a survey of the current situation of multi-object detection and tracking and compares some performances of some visual trackers in order to make some contributions to the development of them. Also, the tracking-by-detection method that runs in every frame is evaluated for the purpose of comparison. The structure of this dissertation is as follows. Firstly, some detection and tracking algorithms are summarized in the literature review. Secondly, some current benchmark datasets for multi-object detection and tracking are presented. Thirdly, an overview of the evaluation metrics for multi-object tracking is summarized. Finally, these multi-object tracking methods are compared and some conclusions and recommendations are made.