Road Traffic Congestion Analysis Based On Aerial Image

Road traffic congestion analys is is a basic and an important part of intelligent transportation system.Urban traffic trunk road congestion analysis is one of the research hotspot.Urban traffic road congestion analysis is one of the research hotspot which is mainly based on the collection of traffic information and the establishment of traffic jam analysis model,however the traffic information collection mainly depends on the various types of traffic detection equipment which are fixed deployed currently.Due to the high maintenance costs and the incomplete coverage of equipment,the traffic congestion problem has not yet been resolved.In recent years,unmanned aerial vehicles(UAV)have been widely used in the military field and have made substantial progress and breakthrough.UAV detection technology have the advantages of easy maintenance,large amount of information processing ability,and no damaging to the ground.Therefore,it is of vital importance and application value that combine UAV and image processing technology to apply them in the field of road traffic jam analysis.In this article,the road traffic jam based on aerial image is analyzed.Vehicle extraction and vehicle tracking carried out against the motion background are in order to better extract the traffic flow parameters for traffic jam analysis.This article uses traffic flow parameters such as evaluation indexes and establishes congestion analys is model based on fuzzy theory to achieve congestion analysis.The main contributions of this article are as follows:(1)The vehicle is extracted in the motion background.Aim at the background motion estimation,the algorithm based on epipolar constraint matching has been used,the corners are determined according to the size of the response function after the corner res ponse function is obtained by Harr is corner detection.And the non-maximum suppression is performed in the neighborhood of the corner.Secondly,based on the gray information of the corner neighborhood as feature descriptor,the match template method is us ed to initial match of the corner points,and then based on the epipolar constraint condition,the RANS AC algor ithm is used to obtain the fundamental matrix,and according to fundamental matrix to remove the incorrect matching to get the correct matching.Finally,according to the correct match set,the RANSAC algor ithm is used again to solve the optimal transmission transformation matrix,so as to achieve the background compensation.In order to extract vehicle foreground module by background subtraction based on the background compensation,a pixel-based gray voting strategy is proposed to extract background.According to the gray change of the pixel in a certain period of time,the gray scale voting histogram is established.If the grayscale changes slowly,the histogram corresponds to the larger the grayscale voting value,and vice versa.To suppress the effect of light,the neighborhood of each pixel based on a certain area size is in turn polled.According to the voting histogram voting value,to select the maximum voting value corresponding to the gray value as the background to complete the background extraction has been used.Finally,to use background subtraction and morphological corrosion techniques,the road vehicles were extracted.Experimental results show that compared with the mixed Gaussian modeling method and the mean modeling method,this article's algorithm is signif icantly faster than the hybrid Gaussian modeling method,besides,the background modeling effect is equivalent to the hybrid Gaussian modeling method,which is better than the mean modeling method.(2)Vehicle tracking algorithm based on improved mean shift is applied.Firstly,the traditional mean shift algor ithm is used to build the vehicle model.On this basis,in order to overcome the interference of the background area to the vehicle tracking,by comparing the color difference between the vehicle area and the background area,a significant feature amount of the different color gradation of the vehicle is calculated to improve the vehicle model which uses a weighted approach to enhance the role of salient features in the vehic le identification process and to weaken the effect of non-salient features.At the same time,the vehicle model updating strategy is introduced,and the model is updated every few frames,so as to improve the tracking performance.Secondly,for multi-vehicle tracking problem,a multi-thread-based detection tracking composite algorithm is used,for the vehicle detection and vehicle tracking independently assigned to the implementation of the thread at the same time,communication is implemented by each several frames of between the threads and correcting tracking results to effectively solve the problem of tracking changes in the total number of vehicles at all times.(3)Traffic congestion analysis is based on FAHP and fuzzy comprehensive evaluation.Based on the vehicle detection and tracking,the parameters such as vehicle flow and speed are detected by image processing technology,and then calculating the time occupancy rate,saturation and average travel speed of traffic flow parameters which are used as the basic index of vehicle congestion analys is,fuzzy comprehensive evaluation method and fuzzy analytic hierarchy process are used to classify congestion degree into three grades.And the membership function is established by the evaluation index,and the traffic condition composite index is calculated based on the weighted membership degree thus achieving congestion analysis.