中文摘要：

探讨了利用动态图像理解技术对ATM进行智能监控，采用全方位视觉传感器（ODVS）实时采集ATM机周围的全景视频数据；从全景视频数据中提取出各活动对象的属性信息，并根据这些属性信息对各活动对象进行识别；通过对各活动对象的跟踪与分析，得到ATM状态、ATM周围相关人员的行为等信息；最后通过行为语义化处理手段对ATM的多种常见的可疑犯罪行为等进行了识别；实验结果表明，对于模拟ATM一些常见可疑犯罪行为的识别方面，系统具有鲁棒性高、安检内容多、监测范围广等优点，为预防ATM机等金融犯罪提供了一种新的手段。

英文摘要：

The intelligently monitoring is discussed for ATM by using dynamic image understanding technology. It uses the omni-directional vision sensor （ODVS） to acquire the omni-directional image and extracts the properties of active objects from the omni-directional images, which are used to recognize various active objects. By tracking and analyzing various active objects to gain the ATM status information and the behavior information of the relevant people around the ATM. Finally, the behavior semantic processing method is used to recognize various common ATM criminal acts. The results show that for the simulation of the common criminal acts＇ recognizing of the entire ATM using process, the system has the advantages of high robustness and multiple security inspection contents and wide range detecting, and provides a new mean for preventing financial crimes.