Vehicle detection and tracking using YOLO and DeepSORT

• 出版年: ISCAIE 2021 - IEEE 11th Symposium on Computer Applications and Industrial Electronics
• 第一著者: 2-s2.0-85107642947
• フォーマット: Conference paper
• 言語: English
• 出版事項: Institute of Electrical and Electronics Engineers Inc. 2021
• オンライン･アクセス: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85107642947&doi=10.1109%2fISCAIE51753.2021.9431784&partnerID=40&md5=b2c45c12a7670baf4c2c3a261d521e6d

Every year, the number of vehicles on the road will be increasing. as claimed by a road transport department (JPJ) data in Malaysia, there were around 31.2 million units of motor vehicles recorded in Malaysia as of December 31, 2019. While as, from the mid-2017, there were around 28.18 million units of motor vehicles recorded in Malaysia. Consequently, accurate and fast detection of vehicles on the road is needed by using the volume of vehicles as valuable data for detecting traffic congestion which then benefits for traffic management. Using the implemented deep learning for vehicle detection, this paper project is using TensorFlow which is platform for machine learning and you only look once (yolo) which is object detection algorithm for real-time vehicle detection. By combining this two and other dependencies with python as programming language, the suggested method in this paper determine the improvement of YOLOv4 latest algorithm compared to the previous model in vehicle detection system. This vehicle detection also uses DeepSORT algorithm to help counting the number of vehicles pass in the video effectively. From this paper, the best model between YOLO model is Yolov4 which had achieved state-of-the-art results with 82.08% AP50 using the custom dataset at a real time speed of around 14 FPS on GTX 1660ti. © 2021 IEEE.