A multiscale YOLOv3 detection algorithm of road scene object#br#
1. SAIC Motor Commercial Vehicle Technology Center, Shanghai 200438， China； 2. School of Automotive and Traffic Engineering, Jiangsu University, Zhenjiang, Jiangsu 212013， China
Abstract： In natural traffic scene, the bounding box sizes of different road targets vary greatly. The existing realtime object detection algorithm YOLOv3 can not balance the detection accuracy of large and small targets and has poor performance in the task. To solve the problems, the feature fusion module of YOLOv3 target detection algorithm was redesigned to realize the multiscale feature stitching. The detection module was improved by adding two extra feature output modules for small targets, and a new multiscale detection method of YOLOv3_5d for road targets was obtained with 5 detection scales. The experimental results show that the average precision of the improved YOLOv3_5d algorithm is 0580 9 on BDD100K dataset, which is 0082 0 higher than that of original YOLOv3. The running speed is 454 frames·s-1, which can meet the realtime requirement.
