Design of real-time vehicle detection based on YOLOv4
Feng Yang, Xingle Zhang, Shuxing Zhang, Chao Li, Haiwei Hu · 2021 International Conference on Control, Automation and Information Sciences (ICCAIS) · 2021
Intelligent Traffic System (ITS), whose key point is vehicle detection, can significantly reduce the incidence of traffic accidents through scientific traffic control. Traditional vehicle detection technology has unsatisfactory detection speed and accuracy. In order to obtain and process object feature information more efficiently, we use the convolutional neural network (CNN) with better performance. In addition, compared with traditional computers with large volume and high computing cost, the embedded platform is more suitable for deployment model in practical application. Based on the convolutional neural network YOLOv4, we deploy it on the embedded platform, and optimizes the Class-imbalance of BDD100K dataset used in the model and the lack of real-time performance of the model. After optimization, for the BDD100K dataset, the detection accuracy of the model is improved by 2.81%, and the detection speed is improved to 28.45 FPS, which can meet the requirements of real-time vehicle detection.