Pedestrian-Vehicle Detection Model Based on Optimized YOLOv3

Vehicle and pedestrian detection has always been an important and challenging task in the autonomous driving technology. To solve this problem, this paper proposes an optimized method based on YOLOv3, which is utilized to detect vehicles, pedestrians and cyclists. On the basis of traditional YOLOv3, this algorithm increases the number of deep residual structure and feature map sizes. And then cascades feature maps of various sizes in the backbone network to obtain the Feature Map-Concat (FC) structure. We also optimize the regression loss function at the same time. The evaluation index of this algorithm was implemented on KITTI data set, the results showed that the proposed detection algorithm proposed in this paper achieved encouraging results.