Research On Abnormal Events Detection In Video Surveillance Of Public Bicycle

As a kind of low-carbon and environmental transport,public bicycles can effectively solve the last one mile problem,and play an irreplaceable role in urban public transport system.However,current public bicycle system runs that one bicycle is equipped with one pile,which leads to serious waste of land resource and costing a lot.Therefore,the public bicycle without pile is one of the trends in the future development.However,without the limitation of the fixed pile,the safety of the bicycle can not be guaranteed,some abnormal events may occur in the parking area,e.g.,in the absence of permission,the public bicycle is moved,moving objects loiter in the parking area,etc.The traditional video surveillance system can not give an alarm when abnormal events occur.Therefore,according to the development needs of a Wuxi public bicycle company,in order to detect the abnormal events in public bicycle parking area,this paper carries out the research of public bicycle abnormal event detection in video surveillance,and the embedded application platform is designed to validate the research results,and provides technical support for the company development of public bicycle system based on machine vision.According to two abnormal events in the video surveillance: public bicycle being moved for no reasons or stolen and moving objects being loitering in the parking area,this paper puts forward two detection algorithm.Finally,the embedded application verification platform is designed,and the two algorithms are transplanted to the embedded application platform.The main works of this paper are as follows:(1)According to high complexity and large amount of noise in the outdoors,a moving object detection algorithm based on an improved visual background extraction algorithm is proposed to detect the abnormal events that the public bicycles are removed by people for no reason.According to the disadvantages of traditional ViBe(visual background extractor)algorithm that the Vi Be algorithm is easy to produce ghosting and poor ability to adapt to the dynamic scenes in the initialization background model,the number of times that a foreground pixel is detected as stopped is recorded to inhibit the ghost and the update threshold is dynamically adjusted according to the scene complexity to improve the adaptability.The experimental results show that the detection precision of the proposed algorithm is improved by more than 30% compared with the ViBe algorithm.(2)In order to detect the abnormal wandering events in the surveillance area effectively,a moving object detection algorithm based on moving object trajectory analysis is proposed.Extract the motion trajectory of the target,calculate the direction of motion of discrete points of trajectory and the principal direction angle of the trajectory,finally,we can determine whether the behavior is wandering behavior according pedestrian's motion distance.The algorithm does not need to establish the sample library compared with the other algorithm,which is simple and effective,and can meet the practical application requirements.(3)In order to verify the research results,this paper designs an embedded application platform.It mainly includes the design of the hardware platform,the establishment of the embedded Linux operating system and the software design.Based on the platform,the results of the above research are verified.The results show that the abnormal events detection in public bicycle video surveillance can be realized in complex dynamic scene to meet the needs of practical application.