Vehicle target detection algorithm based on improved YOLOv3

Abstract

The detection of road vehicle targets is an important research direction in the field of target detection. As there is some problem in vehicle target detection while using YOLOv3, such as high miss rate and low detection accuracy, an improved multi-scale feature fusion target detection algorithm IFFC-YOLOv3 is proposed. Firstly, optimizing the multi-scale feature fusion based on the YOLOv3 algorithm to improve the detection of small targets while ensuring the detection capability of the network for medium and large targets. Secondly, using K-Means++ target frame clustering on road target detection turns out to obtain new candidate frames to improve its accuracy. Finally, the CIoU optimized loss function is introduced to further improve the accuracy of vehicle target detection. In this paper, the effect of the improved IFFC-YOLOv3 algorithm is verified by means of a personal datasets. The experimental results show that the IFFC-YOLOv3 algorithm achieves good detection results in terms of reducing the rate of missed detection and improving the detection accuracy.