Application of YOLO V2 in Construction Vehicle Detection

Based on the characteristics of YOLO network, high efficiency, high regression rate and less computation, this paper proposes the design of on-line detector for construction vehicles. YOLO V2 is selected as the data training network for this time by conducting vehicle detection comparison training for network methods such as CNN, R-CNN, Fast R-CNN and Faster R-CNN in deep learning, and aiming at the requirements of online vehicle detection for anti-blocking construction. By collecting vehicle data on the construction site and using the Video Labeler tool in MATLAB to mark the data set, the vehicle detection training data set was established. By using Resnet 50 network as the feature extraction network, the detector network structure was constructed, and the anchor frame was generated by the central point clustering method, which realized the training of YOLO V2 detector network. Through simulation verification, the designed detection accuracy reaches over 94.79%, which meets the needs of online detection.