The Design And Implementation Of Escalator Passenger Detection And Tracking System Based On TMS320C6748

With the extensive use of escalator in recent years,some security issues aroused more and more people’s attention,such as passenger flow congestion,passenger fall at escalator exit.Artificial care is most commonly used,but it is costly and difficult to monitor in real-time.The rapid development of intelligent video surveillance technology,such as image processing technology and embedded technology,for escalator manufacturers to solve these problems provides a new way of thinking.This paper aims to design and implement a system for the detection and tracking of escalator passenger,offers a possibility for the analysis of high-level passenger’ behavior.With full account of the processing power,scalability,cost-effective and others,the TMS320C6748 was selected as the system main chip for the system hardware platform.And TVP5147,RS485 were used for video acquisition and data communication respectively.According to the characteristics of application scenario,the vertical axis of the monocular camera installation method was adopted.The detection algorithm of this system mainly includes three aspects.1)Passenger target detection.Experiments of common-used moving target detection methods were carried.Based on Hu invariant moments theory,the circularity of passenger’ head was fused to detect passenger target.The theoretical analysis and experimental simulation of HOG + Adaboost were also carried out.Experimental results of the above two methods were compared.2)Passenger target tracking.Based on the common-used moving object tracking methods,this paper analyzes the difficulty of tracking in this application scenes.The moving target tracking and analysis based on Kalman filter was proposed.The matching similarity function was defined,and the adjacent-frame matching multi-target tracking strategy was adopted to realize the tracking of moving objects based on adjacent-frame matching and Kalman filter.This algorithm has a good performance in real-time.3)Large object retention detection.In this paper,a novel detection method based on fast and slow background difference strategy of Surendra was proposed.Two kinds of learning coefficients were chosen to detect the large object retention by using the difference of background update rate.Finally,the detection algorithm were transplanted to the DSP platform and optimized.After the system test and communication test,it is proved that the proposed system can well accomplish its tasks as expected.The research of system is of great significance for the intelligence of escalator operation monitoring and passenger safety protection.