Research On Passenger Flow Counting Algorithm Based On 3D Trajectory Pattern Classification Technology

The passenger counting based on video processing technology has become an important part of advanced public transportation system.The traditional method is to use a monocular camera.Nevertheless,this kind of counting method generally has two aspects: on the one hand,it can’t solve the problem of target occlusion,which is easy to lose the target.On the other hand,it does not make full use of the three-dimensional information of the target trajectory.Therefore,this paper uses the RGB-D camera calibration parameters to transform the two-dimensional trajectory into three-dimensional space.This paper researches on the public transport passenger flow counting algorithm based on three-dimensional trajectory pattern classification technology,and accurately count the public transport passenger flow.In this paper,a three-dimensional trajectory pattern classification technology is used to calculate the public transport passenger flow.The basic view of the algorithm is to detect,track and classify.Firstly,Kinect depth camera is utilized to obtain the depth map.According to the correspondence between the height and the depth value of the depth map,the target of the head is locked based on the restriction of maximum feature of the target.Secondly,the Kalman filter is utilized to predict the position of the head in the next frame.Block matching method is utilized to achieve the matching and tracking suspected head targets.The 2D trajectory is obtained on the image.Through the calibration parameters of the camera,the 2D trajectory in the image coordinate system is converted into the 3D trajectory in the world coordinate system.Three different trajectory classification methods are used to realize the bus passenger flow counting: the classification algorithm based on the detection line,the 3D trajectory classification algorithm based on Adaboost and the 3D trajectory classification algorithm based on SVM.In this paper,three trajectory classification algorithms will be verified.Experimental results demonstrate that the above three algorithms met the real-time requirements.The precision of classification algorithm based on detection line was 90.87%,the accuracy of classification algorithm based on Adaboost was 95.36%,and the accuracy of classification algorithm based on SVM was 96.18%.The detection accuracy of classification algorithm based on the detection line is low and it does not have the versatility.Nevertheless,the algorithm establish on a supervised pattern classification has high accuracy.In comparison,SVM classification algorithm gets the highest accuracy and applicability.