An AI-based recognition system for abnormal human behavior in elevator car
基于AI的电梯轿厢内人员异常行为识别系统

摘要

电梯里因遮挡和光照之类的干扰因素，再加上空间不大，以往用手工设计特征提取的行为识别方式因鲁棒性和太过复杂等等不足并不是太好，而且电梯里有交互行为和个体行为，在人多的时候识别异常行为也较为困难。特别是孩子和女性在乘坐电梯时碰到异常的行为概率更大，而且很难有效发出求救信号和保护自己；老弱病残，若一个人在乘坐电梯时，因疾病或其他原因昏倒在电梯内，也很难短时间内取得有效救助，往往会受到生命危险。通过视频分析，模式识别以及人工智能各类高科技手段，智能视频监控系统可以识别检测监控画面内的人数以及行为是否异常。利用主帧解析视频流取得处于图像真理的目标并捕捉目标的行为与动作，同时剖析视频内容标记并报警出现的异常举动。

Abstract

In the elevator, due to interference factors such as occlusion and light, coupled with a small space, the past behavior recognition method with manual design feature extraction is not very good because of the lack of robustness and too complex, and there are interactive behaviors and individual behaviors in the elevator, and it is difficult to identify abnormal behaviors when there are many people. In particular, children and women are more likely to encounter abnormal behavior when taking the elevator, and it is difficult to effectively signal for help and protect themselves; If a person passes out in the elevator due to illness or other reasons when taking the elevator, it is difficult to obtain effective assistance in a short period of time, and often his life is in danger. Through video analysis, pattern recognition and various high-tech means of artificial intelligence, intelligent video surveillance system can identify and detect the number of people in the monitoring screen and whether the behavior is abnormal. The main frame is used to analyze the video stream to obtain the target in the image truth, capture the target's behavior and action, and analyze the video content marker and alarm the abnormal behavior.