An Edge Computing-Enabled Track Obstacle Detection Method Based on YOLOv5

Rail transit is developing towards intelligence, which requires a lot of computing resources to carry out deep learning tasks. However, the limited computing power of on-board equipment hinders the operation of complex detection networks for the execution of complex deep learning tasks. And the existing railway inspection methods are mostly based on fixed equipment, their detection range is limited. In this article, we propose a mobile platform railway inspection method based on edge computing. Edge computing can reduce stress by offloading the workload to edge nodes. The mobile platform vehicle is responsible for acquiring real-time images. Edge computing servers are used to help execute object detection algorithms off-orbit and carry most of the computing power. The use of edge computing architecture for railway inspection can effectively reduce the pressure of the on-board server and reduce the network dependence. Aiming at the complex scene of track detection, we propose an improved YOLOv5-based target detection algorithm. By adjusting the network model and deploying it on the edge server, the method reduces the computing burden and realizes the collaborative reasoning with the whole edge computing architecture. We implement the collaborative reasoning scheme in practical experiments and find that the proposed improved target detection method based on edge computing can complete the target detection task with few computing resources, and meet the demand for precision and real-time performance.