A Improved Yolov4’s vehicle and pedestrian detection method
Conference: ICMLCA 2021 - 2nd International Conference on Machine Learning and Computer Application
12/17/2021 - 12/19/2021 at Shenyang, China
Proceedings: ICMLCA 2021
Pages: 7Language: englishTyp: PDF
Personal VDE Members are entitled to a 10% discount on this title
Authors:
Wang, Hailong; Tian, Shihe; Zhang, Zhian; Wang, Husen (Nanjing University of Science and Technology, AIR Robot lab, Nanjing, China)
Abstract:
Aiming at the difficulty of existing target detection algorithms to detect vehicles in complex road environments, an improved Yolov4 vehicle detection method is proposed. At the same time, this method distinguishes the front, side, and right sides of the vehicle, and can determine the direction of the vehicle based on the detection results. This method has been improved in the Yolov4 Backbone, combined with a lightweight attention mechanism-Coordinate attention, to improve the ability of network feature fusion, and it can effectively solve problems such as vehicle density. The mAP of the method in this paper reaches 92.2%, which is 5.4% higher than that of Yolo. Compared with other mainstream algorithms, the detection accuracy is greatly improved, which proves the effectiveness of the algorithm.