Key Object Detection and Recognition Method for Abnormal Events in Waiting Halls

In recent years, with the enhancement of the domestic railway network and the widespread adoption of high-speed rail, the issue of safety management within waiting areas has become increasingly prominent. This challenge is characterized by high passenger mobility and diverse behaviors. Traditional manual inspection methods are inefficient and struggle to provide comprehensive coverage. Meanwhile, the application of generic visual technology in complex waiting environments is limited due to the scarcity of anomalous data and class imbalance, leading to difficulties in detection. This paper proposes a method for key object detection and recognition based on object detection techniques. By optimizing single-stage detectors and employing data augmentation strategies, this method addresses the issue of insufficient samples, thereby enhancing the safety management level of waiting areas.