Research And Implementation Of Bus Crowd Counting System

As an important part of the intelligent public transportation system,bus crowd counting system can provide important data support for bus line allocation and vehicle scheduling,which is of great significance to the intelligent management of bus operation scheduling.This thesis uses bus monitoring video data to study bus crowd counting algorithm based on deep neural network and computer vision technology,and implements a bus crowd counting system.The main work of this thesis is as follows:First,this thesis proposes a detection algorithm based on convolutional neural network for door switch state detection.By building a full convolutional classification network,the video images are “opened” and “closed”,and the detection of the door switch state is realized.Through the optimization of the network model,the parameters of the network model designed and implemented in this thesis are very small and run fast.At the same time,because of its full convolutional structure,there is no limit to the input image size.In this data set,the network has achieved very good classification results.Second,aiming at the problem of human head detection in buses,this thesis makes a bus head detection data set based on about 173 hours of real bus monitoring video,and puts forward the method of data labeling combined with pre-training model.On this data set,this thesis optimizes and trains the target detection algorithms Faster-RCNN,RFCN,and SSD algorithms based on deep learning.By testing the detection speed and precision of these algorithms,the network structure of ResNet101+RFCN is selected as the human head detection of this thesis.algorithm.Third,we designed and implemented the crowd counting algorithm under the bus surveillance video scene: For the crowd density estimation under the bus surveillance video scene,this thesis studied and implemented the MCNN algorithm and the crowd density estimation algorithm based on human head detection.The accuracy and speed of the two algorithms are compared and analyzed.For the traffic counting problem in the bus surveillance video scene,this thesis uses the detection information acquired by the human head detection algorithm to obtain the passenger’s movement trajectory through the inter-target target correlation algorithm.Based on the trajectory,a cross-line counting algorithm was designed to realize the counting of passengers getting off the bus.Finally,using of bus crowd counting system-related algorithms,the realization of the bus crowd counting system software design,real-time display of the current status of the bus and the crowd count results to help users understand the relevant information.