Research On Moving Objects Detection Algorithms In Security Monitoring System

With the development of monitoring technology, the intelligent visual surveillance system which put machine vision into monitoring technology has become a new direction. Intelligent visual surveillance technology is involved in machine vision, image processing, pattern recognition, artificial intelligence and many other research fields. It is widely used various application fields and has great potential in market value and social benefit. As the front part of intelligent visual surveillance technology, the moving objects detection is the foundation and key to the following object tracking, recognition and action understanding.According to the actual situation of Security Monitoring System, some common methods in the field of moving object detection, such as Optical Flow method, Frame Difference method and Background Subtraction method are discussed in this paper. The advantage and shortcoming of three algorithms are analyzed and compared. Then according to the characteristics of embedded security monitoring system and the requirement of computational complexity, real-time and accuracy, the single-Gaussian model based on the background subtraction is chose as the moving objects detection method of this system.According to the requirement of security monitoring system, the hardware platform based on ARM7 is designed in this paper. It selects LPC2136 as the main controller and has the function of image acquisition and processing, induction anomalies and GPRS modules for MMS transmission and so on. Meanwhile, as in the case of the first frame in the video sequence contains a moving object and the change of objects statement in the monitoring region, the traditional single Gaussian model cannot create a background image accurately. So an improved single Gaussian model is proposed, and it solves the problem of creating background image in the above-mentioned special cases. The improved method is simulated in the MATLAB. The simulation results show that this improved single Gaussian model can extract the moving object effectively and have a good detection result.