Hardware-Accelerated Real-Time Detection of Catenary Geometry Parameters

Dynamic detection of catenary geometry parameters (CGP) has a significant impact on the stability of the train power supply. Aiming at the problem that the detection speed of traditional detection methods can not meet the high frequency, high speed, and low fitness, this paper presents a non-contact, high-precision, real-time detection system for the CGP based on FPGA hardware acceleration. Dynamic measurement techniques based on monocular cameras and line-structured light. Firstly, the FPGA controls the camera to acquire images in real-time, and a parallel three-step median filtering algorithm is developed to reduce image noise, ten times more efficiently than the traditional median filter. Secondly, in order to solve the catenary localization problem in complex scenes, Hough transform is utilized to detect catenary, and the mean-shift algorithm is improved according to the hardware structure to realize catenary localization and tracking in complex scenes. Additionally, the weighted centroid extraction algorithm combined with hardware acceleration is optimized to achieve the stable extraction of pixel coordinates. Finally, the FPGA migration of the algorithms is completed. The coordinates of the spot center are transmitted to the host via Ethernet, and CGP is calculated by reconstructing the line structured light in 3D. After testing an image with a resolution of 800*480, Modelsim simulation takes 0.02 ms to process one image frame. When the algorithm is transplanted to the FPGA platform, the process from sending the control word to receiving the center of mass coordinates takes only 25 ms. The experiments demonstrate that the detection system outlined in this paper exhibits high accuracy and speed, adaptability, portability, and ease of installation, making it highly valuable for practical applications.