Monitoring the Safety Operations of Escalators Using a Multi-Level Fusion Network

With the increase in railway passenger volumes and the expansion of passenger station scales, monitoring the safe operation of escalators has become increasingly challenging. Traditional manual inspection methods are inefficient and prone to errors, and escalator anomalies can result in serious accidents. Therefore, there is an urgent need for intelligent, automated and precise monitoring of escalator safety. Currently, deep learning, particularly behavior recognition algorithms, has shown outstanding performance in anomaly detection. However, traditional methods fall short of meeting the real-time and accuracy requirements for the complex scenarios of escalators. In response to the unique environment of railway passenger station escalators, this paper proposes an escalator safety operation monitoring model based on a multi-tiered fusion network. This model integrates the perception of elevator operating status, pedestrian movement information, and human key point movement trends to monitor escalator operations and detect abnormal behaviors (such as reverse movement, falls) and luggage drop incidents in real time, thus achieving comprehensive real-time alert functionalities.