Research On Vehicle Target Detection Algorithm Based On Deep Learning

With the continuous improvement of people’s living standards,cars have changed the way people travel,and at the same time bring many negative effects,such as traffic jams,traffic jams and frequent traffic accidents.Under such background,the intelligent transportation system is paid more and more attention.Among them,vehicle detection is an important aspect in the research of intelligent transportation systems,which has great practical significance for solving problems such as traffic congestion,traffic accidents,video surveillance,and license plate recognition.Vehicle detection uses machine learning and image processing technology to determine the sense of certainty Interest vehicle target area.However,the long-distance vehicle target and the overlapping vehicle target occupy less pixel information in the image,the features are not obvious,and they are easily affected by the interference of the lighting environment,etc.,making the existing model unsatisfactory for detection and segmentation effects and obvious obvious such as missing detection.Limitations.Therefore,this paper designs a new vehicle detection and recognition model to solve the problem of low accuracy of long-distance vehicle targets and overlapping vehicle targets in the detection process.The application of vehicle detection technology has become inevitable,and improving the efficiency of vehicle detection is of great significance for maintaining road traffic safety and its management.The main work contents are as follows:Based on the vehicle detection and recognition task flow,this paper improves three vehicle detection and recognition models based on Cascade R-CNN,Mask R-CNN and Cascade Mask R-CNN.In view of the vehicle recognition task’s high requirements on network depth and speed performance,a vehicle detection and recognition network based on improved convolutional neural network is proposed.The improved convolutional neural network is used as the feature extraction network of the three network models of Cascade R-CNN,Mask R-CNN and Cascade Mask R-CNN,in order to improve the recognition accuracy of the model and improve the detection rate of the model again.On this basis,in order to make the information of the vehicle target features more abundant,the region suggestion network is trained to generate candidate windows,and the feature pyramid network is used to merge the detected features to learn the scale feature images of different sizes.Classify and segment the learned images to complete the detection of vehicles on the images.Aiming at the actual scene in this paper,the characteristics of small distance vehicle targets and overlapping vehicle targets,a vehicle detection and recognition network based on improved convolutional neural network is adopted.Improve and optimize its structure on the basis of convolutional neural network: feature extraction network design,improve the detection ability of the network and reduce the number of network layers;candidate window design,the use of bilinear interpolation reduces the region of interest during feature extraction The error condition of the system improves the learning of different vehicle characteristics and improves the accuracy of network detection and recognition.According to the different characteristics of vehicle types,combined with the general environmental conditions(different lighting,shadow occlusion,etc.)and the characteristics of the corresponding three different networks.First select the construction data set,then further process the data set,and use the processed data set to experiment with three different network models under the vehicle images under different lighting and shadow conditions.The experiment shows that the improved convolutional neural network has good structural stability,high detection and recognition capabilities,and a certain improvement in detection rate.The improved Cascade R-CNN network accuracy and detection speed performance is worse than the Mask R-CNN and Cascade Mask RCNN network performance,and the Mask R-CNN and Cascade Mask R-CNN network accuracy and detection speed are different.Not big,and there is a further improvement in speed performance.