Research On Vehicle Detection And Recognition Method Based On Deep Learning

With the development of urban construction and artificial intelligence technology,intelligent transportation system becomes the inevitable choice of intelligent city,which plays a key role in improving road traffic efficiency and reducing traffic accidents.As an important part of intelligent transportation system,vehicle type detection and classification provides data support for intelligent transportation system,which not only facilitates people to travel and grasp traffic information in real time,but also helps traffic police to supervise vehicle violations and quickly track suspected vehicles,greatly reducing police resources and time costs.To solve the problem that traditional vehicle recognition algorithms have low recognition rate due to the influence of shooting distance,light intensity and weather,this paper summarizes and analyzes the convolutional neural network,Goog Le Net network and YOLOv3 network,proposing the separable convolution neural network model and Dense-YOLOv3 model,and completes the detection and recognition of different types of vehicles.This paper focuses on two aspects:1.The traditional vehicle recognition algorithm has the problem of low recognition rate due to the shooting distance,the illumination intensity,the weather and so on.Using the deep learning framework Tensor Flow,basing on the classical Goog Le Net network model,a separable convolution neural network model is proposed to realize the automatic classification of models by adjusting the weight and bias value of superparameters and increasing and decreasing the width and depth of the network.The model is verified by using the BIT-Vehicle standard vehicle data set.The experimental results show that compared with the traditional HOG＿BP algorithm and convolutional neural network model,the separable convolution neural network model has a high recognition rate for equally difficult vehicle images.The average accuracy is 96.30%,which is about 13% higher than that of the ordinary convolutional neural network algorithm,which shows that the model has certain advantages and practical value.2.The traditional YOLOv3 network structure has poor robustness in extracting features such as overexposure or dark light,which leads to low recognition rate.A Dense-YOLOv3 model for traffic vehicle classification is proposed.The model integrates the characteristics ofdense convolutional neural network Dense Net and YOLOv3 network,which strengthening the vehicle model feature propagation and reuse between convolution layers,and which improving the anti-overfitting performance of the network.At the same time,the target vehicle is detected at different scales,and the cross-loss function is constructed to realize the multi-objective detection of the vehicle model.The model is trained and tested on BIT-Vehicle standard data sets.The experimental results show that the average accuracy of the model based on Dense-YOLOv3 vehicle detection reaches 96.57% and the recall rate is93.30%,which indicates the effectiveness and practicability of the model for vehicle detection.