Vehicle Detection Based on Improved Yolov3 Algorithm
Shuai Zhao, Fucheng You · 2020
At present, the degree of urban traffic congestion is increasing, so it is necessary to detect and predict the road traffic situation. However, the current vehicle detection has some problems, such as poor detection effect and inaccurate classification of relatively small vehicles. To solve these problems, an improved YOLOv3 algorithm for vehicle detection is proposed. This algorithm improves the traditional YOLO algorithm. Firstly, it uses clustering analysis method to cluster the data set, and improves the network structure to increase the number of final output grids and enhance the relatively small vehicle prediction ability. Secondly, it optimizes the data set and optimizes the input image. Resolution makes it robust under different external conditions. Experiments show that the improved YOLOv3 algorithm has a higher detection accuracy and a higher detection rate than the traditional algorithm.