The Realization Of Visual Detection And Safe Warning System Of Congestion Degree In Bus

With the development of science and technology,the progress of society and the increase of people’s income,there are more and more vehicles on the road.The increase in vehicles has caused increasing traffic pressure.Faced with the increasingly serious traffic congestion,more and more people choose to take buses in their daily life.Currently,all major cities have basically built a real-time bus location query system to help passengers grasp the real-time location of buses.It is of great significance to further master the crowding degree in buses for passengers’ ride comfort,reasonable bus scheduling and guidance of passengers’ reasonable ride to protect their health under the background of epidemic situation.In this paper,the method of target detection is used to investigate the congestion degree in buses The specific work is as follows:(1)Use the obtained monitoring video in the bus to intercept the relevant video frames,and use the labeling software labellmg to label the vacant seats in the image and the passengers standing in the aisle to make the bus interior detection data set.(2)The classic one-stage target detection network YOLOV3 is used for the detection of empty seats and standing passengers in the bus.The main purpose is to judge the congestion degree and safety warning type in the bus according to the number of empty seats and standing passengers detected in the bus.Due to the complexity of the internal environment of the bus,there are obstructions between the standing passengers and the seats,the accuracy of the final experimental result is low.(3)According to the problems existing in the original Yolov3 network,the network structure was improved.SENet modules were added to each res_unit of Darknet-53,a total of 23 SENet modules were added to improve the feature extraction ability of the network.The original NMS used was replaced with DIOU-NMS and the original mean square error loss function of the network was replaced with CIOU_LOSS,so as to reduce the occurrence of network missed detection.The improved network was trained and tested on the bus internal detection data set,and the test results of the improved YOLOV3 were significantly improved compared with those of the original YOLOV3,with an approximately 9.62%increase in mAP value and an approximately 10%increase in congestion and safety warning judgment accuracy,the accuracy rate eventually reached 85%.(4)Based on the trained model,the bus congestion detection and safety warning system is designed.Select the pictures in the bus that you want to detect,and the system can submit the selected pictures to the trained network model,then according to the congestion degree returned by the network model and the safety warning category,the number of remaining seats in the current bus and the taking suggestions are given.