Software Development Of Escalator Pedestrian Safety Monitoring System Based On Video Analysis

With the rapid development of China’s economy in resent years,escalators have become an increasingly important part of our lives.However,the escalator will keep moving after turned on unless the emergency stop button is pressed.Once the pedestrian has an accident such as a fall,the escalator may cause serious injury to the person.The traditional video supervision system cannot analyze video in real time.In the same time,the manual monitoring method has low recognition rate and slow response speed,and cannot meet the needs of large-scale applications.Therefore,it is of great engineering value to study the pedestrian safety monitoring system applied to the escalator scene.Based on the intelligent,real-time and accuracy requirements of the escalator scene,this thesis designs and develops the escalator pedestrian safety monitoring system software.The system adopts the C/S architecture,which mainly implements functions such as pedestrian fall detection,escalator illegal items detection,and pedestrian retrograde detection.The design is based on openpose multi-person keypoint detection library to acquire the human body structure,and then use the SVM classifier to classify the human body posture,thereby realizing the alarm of pedestrian fall phenomenon in the view.It also uses the SSD target detection framework to obtain targets such as pedestrians and large luggages,then the KCF tracking algorithm is called on the pedestrian target,thereby realizing the alarm of the illegal items and retrograde behaviors in the surveillance video.The video is analyzed in real time,and the screen presents videos as well as results in the form of tiled display wall,which provides sufficient judgment basis for further operation.The test results show that the escalator pedestrian safety monitoring system developed in this thesis has the characteristics of accurate,real-time and high efficiency.In the test environment,the average time required for the single target detection of one picture is about 31.43ms,and the detection accuracy of pedestrian fall and retrograde phenomenon among samples are about 95%and 97%,both can meet system requirements in terms of function and performance.