Traffic Flow Detection Based On Aerial Video And Image

Traffic problem is a common problem all over the world,and traffic flow detection is a very important part of intelligent transportation system.At present,traditional traffic flow detection methods have some shortcomings,traffic flow detecting based on video sequence is one of the main research directions of traffic monitoring.This paper mainly studies using the flexible UAV platform for aerial photograph,and using the vehicle detection and tracking algorithm of the aerial video image to realize the traffic flow detection.In view of the video images collected by UAV aerial photography acquisition system in several different parts of North China,a traffic flow detection algorithm based on UAV aerial photography is proposed in combination with the current traffic flow monitoring requirements.Image enhancement,rotation correction and other preprocessing are carried out to solve the problems of aerial image quality.In order to reduce the search area and error of the large field aerial image detection and tracking,a segmentation algorithm based on HSV color space and a region growth algorithm based on road texture feature is proposed to segmentation the road.The Calman filter estimation and the template matching based on Pyramid are proposed to achieve the vehicle tracking and counting for eliminating the inter frame counting problem and reducing the amount of computation cost,finally the traffic flow of a road is obtained.The experimental results show that the accuracy of the algorithm is88.55%.Based on the floating vehicle method,a traffic flow detection algorithm of aerial image based on floating car method is proposed.The data acquisition process of the floating vehicle method is replaced by rotor UAV aerial photography of the same road several times.Using vehicle detection and tracking counting algorithm based on Calman filtering and template matching in order to judge vehicle motion state in the floating vehicle method based on the aerial image.The experimental results show that the accuracy of the algorithm is 87.7%compared with that of road section traffic flow.By taking the same road to replace the ground floating car method data acquisition process,the vehicle detection and tracking counting algorithm which matches the Calman filter and the template is real.Based on aerial image,the motion state attribute judgment of floating vehicle method is carried out.The experimental results show that the accuracy of the traffic flow detection algorithm based on the floating vehicle method is 87.7% compared with the road section flow.An aerial vehicle detection algorithm based on Faster-Rcnn network is put forward for the problem of aerial image vehicle detection caused by illumination change,brightness and shadow.Due to the small amount of image data taken by UAV aerial photograph system and the unbalance of vehicle categories,before the PASCAL VOC data set is annotated and used for model training,the vehicle image data set should be amplified.Considering the characteristics of the vehicle in the aerial image,the Resnet50 network is used to extract the vehicle features and generate the vehicle candidate regions.Finally,the classification and regression algorithm is used to realize vehicle classification and detection.The experimental results show that the average detection accuracy of training model is 93%.