Research on Vehicle Object Detection Based on Deep Learning

In this paper, deep learning technology is used to study and optimize the vehicle target detection algorithm in the driving environment, and a vehicle target detection model with both real-time and accuracy is constructed. A new model is designed based on MobileNet network, and the DarkNet 53 network structure of YOLO algorithm is replaced and optimized by using deep separable convolutional neural network. The deep separable convolution is used to replace the ordinary convolution layer in the original network structure to reduce the number of parameters in the network structure and improve the learning ability of the convolutional neural network. The corresponding loss function is designed for the new model proposed in this paper. The new model target detection proposed in this paper is studied as a regression problem. Therefore, the design of the loss function comprehensively considers IoU error, coordinate error and classification error to achieve the purpose of accelerating the convergence speed of the model. The new model proposed in this paper is trained on the DeGAN extended KITTI dataset, and the experimental results are compared and analyzed. The experimental results show that the MobileNet-Yolomodel proposed in this paper has different degrees of improvement in model storage and detection speed compared with other algorithms.