Research On Passenger Detection And Counting For Getting On And Off Buses Based On Video Images

At present,China is developing a "bus city",and bus passenger flow information as reference data for bus companies and passenger is very important.With the development of image processing technology,it has become a current research trend to count passenger flow with the help of monitoring video of passengers getting on and off the bus.However,under the influence of illumination and passenger shapes in complex bus scenes,the detection and tracking of passengers is an urgent problem to be solved to realize passenger statistics.The specific research contents of passenger statistics based on video image detection are as follows:(1)A head target detection model based on machine learning is established.First,the head target image samples are collected through various ways,and the HOG features of the samples and the entropy,energy,contrast,and correlation features in the texture features are extracted to form the head target feature database.Secondly,based on the SVM algorithm training features to obtain the HOG feature and texture feature head target classification model,using the HOG feature and texture feature head target classification model classification and recognition accuracy rate is about 86%,which is better than using the HOG feature head alone Target classification model.Then,based on the HOG feature and texture feature head target classification model,head target detection model based on SVM is proposed to realize the detection of passenger head targets.(2)A head target detection model based on deep learning is established.First,the standard of the head target training data set is completed.Then on top of this,after 15,000 iterations of model training,the YOLOv3 head target detection network model is obtained,with a recall rate of 92.12% and an accuracy rate of 89.71%.And it is proved by experiments that the inspection speed and accuracy of the YOLOv3 head target detection network model are superior to the head target detection model based on SVM.(3)A multi-target tracking algorithm based on Cam-shfit and YOLOv3 is proposed.First,the Cam-shfit algorithm is used to track the head target.Secondly,in order to solve the problems such as drift during Cam-shfit algorithm tracking,the tracking effect is optimized by combining the head target tracking data with YOLOv3 detection data through the data association matching method based on the minimum distance.Then,in order to optimize the problem of misdetection and missed detection in the process of head target detection and tracking,combined with time constraints,a rule for judging passenger position information is proposed to improve the reliability of passenger trajectory tracking.(4)A counting algorithm for detecting passengers getting on and off the bus is proposed.First,the passenger trajectory of the bus getting on and off area is analyzed,and the process of judging passengers behavior is proposed.Then,combined with head target detection,tracking and passengers getting on and off behavior judgment research,a counting algorithm for passenger getting on and off is proposed.Finally,carrying out the experiment in the actual bus scene and the simulation scene,the experiment proves that the counting algorithm for the detection of passengers getting on and off the bus proposed in this paper has good detection,tracking and counting effects in the bus scene and the simulation scene.