Study On Moving Object Detection Method For Intelligent Video Monitoring Systems

Intelligent video monitoring system has been an increasingly important research direction in computer vision area and has been investigated extensively by researchers in the world, which is the important high-tech measures and technology to guarantee the public security in modern society. Separating foreground from background in a video sequence is one of the most fundamental tasks in many applications of computer vision, such as video surveillance and monitoring systems, intelligent transportation, intrusion surveillance, airport safety and so on. However, the current moving objects detecting methods are complex, time-consuming, and less robust to the scene variations so that they heavily hinder the practical application process. In order to improve the robustness and the real time ability of intelligent monitoring system and to realize unmanned safeguarding goal, this paper mainly concentrates on the key technique of moving objects detection and shadow elimination for intelligent video surveillance systems. Aiming at this goal, we have accomplished the following research works.1. Moving objects detectionTo detect moving objects, each incoming frame is compared with the background model learned from the previous frames to divide the scene into foreground and background. Therefore, background modeling has been actively investigated in the past decade. The difficulty encountered in background modeling is that the outdoor backgrounds are usually non-stationary in practice. Adaptive Gaussian mixture models and nonparametric models are two popular methods for background modeling at present. However, they are usually too costly to perform for real time applications, since they are both memory and computationally inefficient. To overcome this problem, this paper presents a new method for modeling dynamic background based on clustering theory. For a dynamic background, the histogram of each pixel value over time is usually in the form of multimodal. Therefore, regarding each peak as a cluster, we employ clustering technique to construct and update the model of a dynamic background. Then by using the established background model, the moving objects are segmented from the background quickly and accurately. Moreover, we also describe a post processing method based on mathematical morphology and run-length technique. Experimental results show that the proposed method can effectively capture and adapt to the changes in background, and the moving objects detection method is easy to be implemented for DSP or FPGA based hardware.2. Shadow elimination for moving objectsShadows detection and elimination of moving objects is essential to the post-processing such as objects tracking, classification and recognition. The existence of shadow will allow the above-mentioned post-processing to fail. This paper presents a new shadow distinguishing and eliminating approach by combining edge information and color information in HSV color space in which the run-length connected region technique has been employed to identify and remove the shadows. We have conducted many experiments to verify the proposed approach. The results show that the approach not only remains the integrality of the moving objects, but also meets the real-time requirement.