Abstract: Aiming at the problem that the detection accuracy of the video vehicle recognition method is not high, a YOLOv3 vehicle recognition method based on GIoU is proposed. The Darknet-53 pre-training model was used to carry out transfer learning on the samples of the experimental target vehicles, and the GIoU was used instead of the traditional IoU evaluation method for training. The detected vehicles were divided into two types: buses and cars. Experimental results show that, compared with YOLOv3 trained by traditional IoU evaluation method, the mAP of vehicle recognition is increased by 15%.