Research Of Embedded Railway IntrusiveObstacle Detection System Based On Machine Vision

With the rapid development of China’s high-speed rail, how to ensure operational security has become critical. Among many factors that affect the safety of high-speed railway traffic, obstacle intrusion has become increasingly prominent with higher train speed and increasing braking length, while relying on driver observation and manual inspection have been unable to meet the operational requirements. Consequently, timely and accurately detecting and warning for obstacle intrusion with fixing the specialized devices in key section along the railway play the essential roles.Thesis presents the railway route obstacle intrusion detecting system based on vision machine and embedded technology. Using the embedded hardware platform with ARM as carrier as well as a set of obstacle detecting algorithm specialized on railway environment as core, thesis designs the whole system which is consist of numerous self-detection distributed nodes and layouts them along route to realize the real time detecting function.Thesis firstly designs the hardware detecting platform based ARM, with FPGA and image processing chip to realize image capture and pre-processing function. It accomplishes Ethernet data transmission with transplanting embedded TCP/IP protocol LwIP and a hardware net access. SD card data storage function relies on FatFS file system and SSI communication protocol in ARM. Meanwhile, it designs a set of communication project using rich GPIO ports. Also the experiments show that hardware platform can complete the whole process from image capture to data transmission by Ethernet.An embedded algorithm of obstacle intrusion detection combining obstacle classification with moving behavior analysis is deeply researched. A two-stage discrimination structure is adopted in the algorithm. Firstly, the objects acquired from background subtraction images are classified with Support Vector Machine and a group of eigenvectors, and the most of the moving train are removed from the detection results. Then a Kalman-filter tracking algorithm for left objects is employed, by which the behavior and moving trend of the objects can be analyzed. The analysis results can remove the interference information and improve the alarm accuracy, also early-warning can be done for the obstacle with the intrusion trend.Through many experiments at real railway route, the system’s obstacle warning accurate reaches95.31%, the wrong warning rate and missing warning rate both are below5%, and the average detecting frequency gets13fps. The experiment results show that railway obstacle intrusion detecting system based on machine vision designed by thesis has an acceptable obstacle detecting and warning function, the accurate and real time performance are in line with the system requirement, dangerous target tendency pre-warning also makes a broader range of applications..