Anomalous Event Detection Based on Self-Organizing Map for Supermarket Monitoring

Abstract

Understanding of human behavior is a high-level topic in visual surveillance. This paper brings some traditional algorithms and insights together to construct a framework for a new field called Supermarket Monitoring. Unlike the common surveillance system, the whole body was tracked. In our project, only the moving hands are considered. To fulfill the automated monitoring task, the self-adaptive background subtraction technique and the YIQ skin color model are combined to detect the moving hands. In order to accurate localization at palm, a new method is developed. After successfully detecting the moving hands, a linear prediction model is cited to realize the object tracking. In the behavior recognition stage, the Self-Organizing Map (SOM) is used to distinguish normal behavior from abnormal ones by analyzing the trajectory characterizations of the moving hands. The experiment results show that these methods we advocate are robust and effective. The supermarket should be monitored with special intelligent machine automatically.