A Railway Similarity Multiple Object Tracking Framework Based on Vehicle Front Video

Real-time awareness of the service status of facilities and equipment in metro tunnels is essential for the safe movement of trains. For low illumination underground tunnel environments where objects have high similarity characteristics and are difficult to distinguish, this paper proposes a framework for tracking high similarity objects along the rail line based on on-board visual perception devices. The framework mainly consists of two stages: object detection and object association. In the first stage, an object detection algorithm is used to detect potentially faulty objects in the video, and this paper takes the continuous occurrence of insulated steel supports on the line as the high similarity research object. In the second stage a matching strategy is used to associate the same object in the video sequence and assign the same ID (identity document). Finally, the proposed framework was tested on metro tunnel video and achieved an accuracy of 95.64 and a detection performance of 25 FPS (frames per second), 82.17 MOTA (multiple object tracking accuracy) and 88.46 MOTP (multiple object tracking precision) tracking performance.