Detection System of Truck Blind Area Based on YOLOv3

This letter presents a blind area detection system of the truck based on YOLOv3 algorithm, aiming at the problem that the driver cannot fully observe the surrounding environment because of the multi-directional blind area of the truck. Firstly, the system modified the loss function and the number of anchor frames to realize multi-scale detection. Then it also improved the feature extraction network, the detection accuracy and speed has been greatly improved. In addition, the system also uses a residual neural network as the feature extraction layer to make the prediction output module faster. Finally, Yolov3 draws on the idea of feature pyramid network to predict multi-scale feature graphs, that is, at three different scales, each cell on each scale will predict three boundary boxes. The simulation results show that the blind spot detection system based on YOLOV3 can achieve real-time detection effect, and improve the accuracy of blind spot detection on the condition that the detection speed is maintained.