Publication: On road motorcycle detection and tracking using yolov4
| datacite.subject.fos | oecd::Engineering and technology::Electrical engineering, Electronic engineering, Information engineering | |
| dc.contributor.author | Wong, Li Han | |
| dc.date.accessioned | 2024-02-22T07:55:07Z | |
| dc.date.available | 2024-02-22T07:55:07Z | |
| dc.date.issued | 2021-07-01 | |
| dc.description.abstract | Intelligent transport system is one of the main concepts in smart city which connect mobility, safety, and sustainable living. Intelligent traffic surveillance system for monitoring motorcycles is challenging and rare research in the field. In addition, CCTV captured videos often have poor quality due to cheap sensors and poor ambient lighting condition of the roads. This project aims to build a DNN-based on-road motorcycle detection and tracking framework. Besides, image enhancement techniques as pre-process and Kalman filter-based post processing are used to improve the detection performance. Three Penang roads’ CCTV video datasets are used. The detection algorithm used is YOLOv4 as the speed and accuracy are proven. However, CCTV video datasets have several challenges for detection which includes poor ambient lighting and small motorcycles size. Image enhancement techniques include CLAHE, and super resolution are applied to improve detection performance on CCTV video datasets. Besides, motorcycles that are easily detected in one video frame are not easily detected in other video frames due to occlusion and motion blur. Post-processing includes data association and Kalman are applied to tackle the problem and realise detection-based tracking (DBT). Data association with constraints of smallest Euclidean distance, adaptive gating, and maximum disappearance are used to establish relationship between tracking motorcycles and detected motorcycles for DBT. The final performance of Video 1 improves from 16.40% to 18.03% in average precision (AP), Video 2 has improvement of AP from 0.87% to 1.22% and Video 3 has improvement of AP from 7.08% to 12.89%. after applying pre-processing and post processing proposed in this project. | |
| dc.identifier.uri | https://erepo.usm.my/handle/123456789/18437 | |
| dc.language.iso | en | |
| dc.title | On road motorcycle detection and tracking using yolov4 | |
| dc.type | Resource Types::text::report | |
| dspace.entity.type | Publication | |
| oairecerif.author.affiliation | Universiti Sains Malaysia |