Research On Crowding Detection In Bus Based On Video Images

In the public transportation system of cities in our country,the bus has become the main part of transportation by virtue of its economic convenience and strong carrying capacity.However,the problem of crowded passengers in bus is widespread,which has become an important factor affecting passenger satisfaction.Therefore,it is of great significance for intelligent bus scheduling and improving passenger satisfaction to obtain the information of crowding in bus accurately and timely.With the continuous development of computer vision technology and the gradual popularization of video monitoring system in bus,it is possible to use video images in bus to detect crowding.But there are still few related studies and lack of effective detection technology.Therefore,this paper studies the crowding detection based on video images in bus,the main research contents are as follows:(1)Based on the evaluation standard of crowding in bus,the image sample database is established.Firstly,the standards of coarse classification and fine classification are defined respectively.In the coarse classification,the crowding is divided into comfortable,normal,crowded and very crowded.In the fine classification,the crowding is divided into six grades of A? F;Then,the image samples in bus are collected from video surveillance system,mobile phone shooting and network resources,and a sample database of 16346 sample images is constructed;Next,the sample images is preprocessed by image gray processing,histogram equalization and image denoising to simplify the data and enhance the useful information in the image,so as to improve the effect of the subsequent image feature extraction and recognition.(2)This paper proposes a method for extracting texture features of images in bus combining local binary pattern(LBP)and gray level co-occurrence matrix(GLCM).Firstly,the texture features are selected as the image features based on the analysis of the features of images in bus;Secondly,LBP operator is used to calculate the sample images in bus to get LBP images;Then,the gray level co-occurrence matrix of LBP image in three directions is calculated and the energy,contrast,entropy and inverse difference moment are calculated to form 12-dimensional feature vector.(3)A rough classification model of crowding based on parameters optimization of support vector machine is constructed.Firstly,a rough classification algorithm based on support vector machine is designed;Then,three kinds of SVM parameter optimization algorithms are studied and compared in order to improve the classification accuracy,and the best genetic algorithm is selected as the parameter optimization algorithm;Then,a rough classification model of crowding based on GA-SVM is constructed and verified by experiments.The accuracy of rough classification is 93.20%,and that of fine classification is 78.13%.(4)A fine classification model of crowding in bus based on Res Net is constructed.For the problem that SVM is not ideal in the fine classification of crowding,this paper applies the Res Net model to fine classification of crowding in bus,and construct a fine classification model of crowding in bus based on Res Net-18.The cross entropy loss function is chosen as the loss function and three optimization algorithms are used to optimize.The test results show that Adam algorithm has the best optimization effect and the accuracy reaches 96.22%.Its accuracy is obviously better than that of GA-SVM,which can effectively solve the fine classification of crowding.