Real-time detection of track fasteners based on object detection and FPGA

An accurate and fast fasteners detection is of great significance for improving the inspection efficiency of railway tracks. However, the task is also challenging due to the limited memory and processing capacity of the embedded maintenance systems. To address these challenges, in this paper, we presents a solution that can be deployed on the high-speed operation vehicles, including You Only Look Once version 2 (YOLOv2) network and Field Programmable Gate Array (FPGA) HW-based acceleration strategy. The algorithm proposed in this paper is implemented completely on FPGA, and the hardware acceleration strategy such as loop pipelining, array partition and ping-pong buffer optimization is applied to get the highest performance. The hardware implementation results show that the improved YOLOv2 detection network and fastener detection system are superior to the existing embedded maintenance system in processing time and hardware resources. The proposed systems can detect a track image in 41 ms, which is much faster than the existing embedded maintenance system.