Research On Vehicle Object Detection Based On Deep Learning

Vehicle object detection technology is one of the key technologies for autonomous driving.It requires real-time and accurate acquisition of vehicle information around the vehicle body to improve the driver’s safety.The current stage of the vehicle object detection algorithm based on deep learning can be divided into Two-stage detection algorithm and One-stage detection algorithm.The Two-stage algorithm has high detection accuracy,but the detection speed can not meet the real-time requirements.The One-stage detection algorithm basically meets the real-time requirements,but the detection accuracy is low.As a one-stage detection algorithm commonly used in the industry,YOLOv3 has good detection accuracy as well as real-time.However,when it is applied to the vehicle detection scene,it still lacks real-time and accuracy.In view of this,this paper improves the YOLOv3 algorithm and applies it to vehicle object detection to solve the problems existing in vehicle object detection.The main work of this paper is as follows:(1)Aiming at the problems of high memory usage and insufficient border positioning when the YOLOv3 algorithm is applied to vehicle object detection,a vehicle detection algorithm based on Shuffle Net is proposed.Firstly,in order to reduce the number of parameters and compress the size of model memory,the efficient convolutional neural network structure in Shuffle Net is used to light-weight the Dark Net-53,so as to improve the detection speed.Secondly,in the data preprocessing stage,an improved K-means algorithm is proposed to cluster the labels of the data set to obtain a more suitable anchor boxes,thereby improving the accuracy of border positioning.Finally,in the model training stage,Focal Loss is used to improve model confidence loss,and CIo U Loss is used to improve model regression loss,in order to improve the learning ability of the model,thereby improving the detection accuracy.Experimental results show that on the KITTI dataset,compared to the YOLOv3,the memory and parameters of the improved model are less than half of the original model,the detection speed is 1.5 times that of the original model,and the m AP is increased by 2.12%.(2)Aiming at the problems of poor detection effect of feature layer and easy missed detection of overlapping objects when the YOLOv3 algorithm is applied to vehicle object detection,a vehicle detection algorithm based on a Multi-Level Feature Fusion Network is proposed.By improving the YOLOv3 feature fusion network part,a Multi-Level Feature Fusion Network structure is proposed to improve the detection effect of the feature layer.In the detection stage,a Soft Non-Maximum Suppression algorithm is used to filter the prediction box to reduce the occurrence of target missed detection.Experimental results show that on the KITTI dataset,compared with the YOLOv3,the improved model maintains the same detection speed,and the m AP is increased by 4.91%.