Motion Control System of Unmanned Railcars Based on Image Recognition

The main purpose of this paper is to construct an autopilot system for unmanned railcars based on computer vision technology in a fixed luminous environment. Four graphic predefined signs of different colors and shapes serve as motion commands of acceleration, deceleration, reverse and stop for the motion control system of railcars based on image recognition. The predefined signs’ strong classifiers were trained based on Haar-like feature training and AdaBoosting from Open Source Computer Vision Library (OpenCV). Comprehensive system integrations such as hardware, device drives, protocols, an application program in Python and man machine interface have been properly done. The objectives of this research include: (1) Verifying the feasibility of graphic predefined signs serving as commands of a motion control system of railcars with computer vision through experiments; (2) Providing reliable solutions for motion control of unmanned railcars, based on image recognition at affordable cost. The experiment results successfully verify the proposed methodology and integrated system. In the main program, every predefined sign must be detected at least three times in consecutive images within 0.2 s before the system confirms the detection. This digital filter like feature can filter out false detections and make the correct rate of detections close to 100%. After detecting a predefined sign, it was observed that the system could generate new motion commands to drive the railcars within 0.3 s. Therefore, both real time performance and the precision of the system are good. Since the sensing and control devices of the proposed system consist of computer, camera and predefined signs only, both the implementation and maintenance costs are very low. In addition, the proposed system is immune to electromagnetic interference, so it is ideal to merge into popular radio Communication Based Train Control (CBTC) systems in railways to improve the safety of operations. View Full-Tex