Research On Detection And Pedestrian Recognition Of No Signal Zebra Crossing

According to Article 47 of the road traffic safety law of the People’s Republic of China,when a motor vehicle passes through a crosswalk,it shall slow down.Zebra crossing and pedestrian detection technology is particularly important for auxiliary driving system and unmanned driving system.Aiming at the problem of pedestrian detection at zebra crossing in the environment of no signal,this paper proposes a semantic segmentation about zebra crossing detection and the deep learning about pedestrian detection.The specific research contents are as follows:In the aspect of zebra crossing detection,this thesis realizes the recognition of zebra crossing based on the improved semantic segmentation algorithm.The mobile net is used to replace the coding part of segnet network as the main feature extraction network.The features of the image are extracted by down sampling,and the feature layers of different sizes are obtained.Then the feature layer is decoded by up sampling through the decoding part of segnet network,Finally,the softmax function is used to classify the pixels and complete the image segmentation.The experimental results show that the improved algorithm can greatly reduce the time of zebra crossing detection on the premise of ensuring the accuracy of zebra crossing detection.In the aspect of pedestrian detection,pedestrian detection is realized based on the improved YOLOv3 algorithm.The improvement is divided into the following aspects: firstly,the shallow feature output layer is added at the sampling point of quartic the feature extraction network,so that the network can detect pedestrian objects of different sizes on four size feature maps;Spp module is added to fuse the local and global characteristics of the image to improve the accuracy of model detection;The giou frame loss function is used to replace the IOU loss function to improve the accuracy of frame prediction;Finally,the new prediction anchor frame is generated by kmeans clustering algorithm for network training.The experimental results show that the improved YOLOv3 algorithm improves the effect of small target pedestrian detection,and the detection accuracy has been significantly improved.Finally,the contour of the image is recognized based on Canny algorithm,and the region of interest is extracted.The intersection ratio of pedestrian area is detected by calculating the region of interest and the improved YOLO algorithm,and the pedestrian on the zebra crossing is selected.The experimental results show that the pedestrian detection system on the zebra crossing has achieved satisfactory results,which has a positive significance for the safety of driving in life,and provides the basis for the driverless vehicle in the traffic conditions without signal lights.