基于全方位视觉的ATM机智能监控研究

发布时间：2018-04-23 03:34

本文选题：全方位视觉 + ATM机智能监控　；参考：《浙江工业大学》2009年硕士论文

【摘要】： 近年来,自动取款机(ATM)在各商业银行大量投入使用,在方便用户存取款的同时,也带来了越来越多的ATM机纠纷案件和金融犯罪,而目前的银行系统由于缺乏智能化监控手段导致此类犯罪案件持续上升。本文将全方位视觉传感器引入ATM机智能监控领域,克服了现有监控系统监控范围小、智能化水平低、图像信息融合困难等缺陷。本文的主要工作和研究成果如下: 1.设计了一套基于全方位视觉传感器的ATM机智能监控装置并用Java实现了一套完整的ATM机智能监控系统,该系统可以完成ATM环境中的徘徊滞留行为、窥视跨线行为、敲砸暴力行为和张贴虚假广告行为的识别,实验证明该系统实时性和鲁棒性都比较理想。 2.在前景目标提取方面,针对所关心的不同对象研究了混合高斯算法、肤色人脸检测算法、最大熵阈值分割算法和梯度纹理分割算法,在对上述算法详细分析和研究的基础上分别进行了代码实现。对混合高斯算法做了改进以适应ATM环境中用户时停时走的情况,对不同色彩空间和肤色模型下的人脸提取效果做了实验比较。 3.在运动目标跟踪方面,分析了现有跟踪算法的优缺点,对其中应用广泛的颜色跟踪算法(CAMSHIFT)和交互式多模型跟踪算法(IMM)做了深入研究和代码实现,对这两种算法的跟踪效果做了实验比较。 4.在高层行为语义理解方面,在分析了目前常见的行为理解方法及优缺点的基础上,将支持向量机(SVM)算法应用于人体轨迹训练和徘徊滞留行为识别,将行为语义规则判定应用于其他三种异常行为的识别。在用户使用ATM机的同时抽取出用户人脸图像,以减少存储负担和方便后续的人脸识别等。
[Abstract]:In recent years, ATM (ATM) has been put into use in a large number of commercial banks. While it is convenient for users to deposit and withdraw money, it has also brought more and more ATMs dispute cases and financial crimes. The current banking system, due to the lack of intelligent monitoring means, such crimes continue to rise. In this paper, the omnidirectional vision sensor is introduced into the intelligent monitoring field of ATM, which overcomes the defects of the existing monitoring system, such as small monitoring range, low level of intelligence, difficulty in image information fusion and so on. The main work and research results of this paper are as follows: 1. A set of ATMs intelligent monitoring device based on omni-directional vision sensor is designed and a complete ATM intelligent monitoring system is implemented with Java. The system can realize hovering behavior in ATM environment and peep across lines. The experiments show that the real time and robustness of the system are ideal. 2. In the aspect of foreground target extraction, mixed Gao Si algorithm, skin color face detection algorithm, maximum entropy threshold segmentation algorithm and gradient texture segmentation algorithm are studied for different objects concerned. On the basis of the detailed analysis and research of the above algorithms, the code is implemented separately. The mixed Gao Si algorithm is improved to adapt to the situation of user time stoppage in ATM environment, and the effect of face extraction under different color space and skin color model is compared experimentally. 3. In the aspect of moving target tracking, the advantages and disadvantages of the existing tracking algorithms are analyzed. The color tracking algorithm (CAMS HIFT) and the interactive multi-model tracking algorithm (IMM), which are widely used, are studied and implemented in code. The tracking effect of these two algorithms is compared experimentally. 4. In the aspect of high-level behavior semantic understanding, based on the analysis of common behavior understanding methods and their advantages and disadvantages, support Vector Machine (SVM) algorithm is applied to human trajectory training and hovering behavior recognition. The behavior semantic rule decision is applied to the recognition of the other three abnormal behaviors. In order to reduce the storage burden and facilitate the subsequent face recognition, the user face images are extracted simultaneously by using ATM machine.
【学位授予单位】：浙江工业大学
【学位级别】：硕士
【学位授予年份】：2009
【分类号】：TP277

【引证文献】