Train-to-Edge Cooperative Intelligence for Obstacle Intrusion Detection in Rail Transit

The state of the rail track and its surrounding environment have a significant impact on train operation. The presence of any unusual obstacles on the rail track can pose a serious safety risk to rail transit. Accurate track-side obstacle intrusion detection can assist trains in positioning correction and improve train autonomous operation capability. Existing intelligent obstacle intrusion detection approaches ignore the onboard resource constraints, and the detection model does not consider the practical rail transit environment. This paper proposes a train-to-edge cooperative computing framework for track-side obstacle intrusion detection. We propose a deep learning model consisting of two stages, which allows for the detection of obstacle intrusion and the calculation of the distance between the train and the obstacle in a lightweight manner. In order to reduce detection time, we have developed a multi-agent reinforcement learning model that formulates the task of offloading the detection model inference and migrating edge computing tasks. Our experimental results demonstrate that the two-stage intrusion detection model can achieve higher accuracy, and our proposed collaborative train and edge computing framework can offer real-time computing services to the train.