Design of inspection system for railway engine room based on visual detection

An inspection system was designed to deal with the issue of manual inspection in railway engine room. The inspection system consists of robot layer, cloud server layer and client layer. The robot detects the environment of room and cabinet status through various sensors, the detection results are sent to the cloud server, and shown in client software. Considering the complexity of cabinet status indicators, PVAnet was applied to the cabinet detection. The videos of the cabinet were collected for classification, labeling and training. The neural network model was tested online and the final result proved the effectiveness of the proposed algorithm.