Key-point based tracking for illegally parked vehicle detection

Gao, Xing (2021) Key-point based tracking for illegally parked vehicle detection. Doctoral thesis (PhD), University of Sussex.

[img] PDF - Published Version
Download (5MB)

Abstract

This research aims to develop a target detection and tracking system that can realize real-time video surveillance. The purpose of the research is to realize a monitoring application that can run automatically and intelligently to detect and track illegally parked vehicles. Since the application scenario of the algorithm is a real traffic environment, it must be able to adapt to complex environmental interference, such as drastic changes in lighting conditions, frequent occlusion, and long-term stable tracking.

The thesis shows the detailed design process and test results of the system. This algorithm combines the target detection function based on deep learning network and the multi-object tracking algorithm based on key point matching. The method shown in the thesis focuses on detecting and tracking stationary vehicles in the no parking area. An object detection algorithm based on a deep learning network is used to recognize vehicles. Once the recognized vehicle is defined as an illegally parked vehicle through the determination of its motion state and location, an algorithm based on key-point matching is developed and tracked for this type of vehicle. If the target is still stationary in the no parking area after a period, the system will generate an alarm.

The method was tested in more than 20 hours of video. The video comes from public database and our own. They all show real surveillance scenes, including different time periods of the day and different locations. The test results show that the method achieves 100% in precision (also called positive predictive value), 95% in recall (also known as sensitivity) and 97% in F1 (a measure that combines precision and recall). The results obtained also produce better detection and tracking compared to other comparable methods.

Item Type: Thesis (Doctoral)
Schools and Departments: School of Engineering and Informatics > Engineering and Design
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7800 Electronics > TK7880 Applications of electronics
Depositing User: Library Cataloguing
Date Deposited: 17 Nov 2021 12:26
Last Modified: 17 Nov 2021 12:26
URI: http://sro.sussex.ac.uk/id/eprint/102982

View download statistics for this item

📧 Request an update