A Robust Detection Method For Moving Targets Based On Frame Difference

Moving target detection technology is an oncoming application and concerned leading topics in the field of computer vision.With the rapid development of the network and the digital video technology, the tendency of machine vision towards to the intellectualization and cyberization. The purpose of machine vision is how to make machines obtain visual processing capabilities similar to human beings, with which machines are able to assist or even replace human work. Moving target detection technology is not only the key issue of computer vision, but also a necessary process of acquiring dynamic visual information.Almost all visual monitoring systems begin with the moving target detection, whose purpose is to partition the regions related to moving targets from others. With strong theoretical and practical significance, the research of moving target detection has a wide spectrum of promising applications.In recent years, moving target detection technology play an important role in many branches of science and engineering applications, thus have been received much attention. Some important algorithms were proposed in this research area to solve some practical problems. However, notice that the most existing methods are contrapose and applicable to a particular background, and can achieved good results for some prominent factors in a target background, but can not completely solve the impact of a complex target background. Therefore, there still are some problems left in the moving target detection, and need to be solved.The practical environment of the visual system is very complex, thus it is urgent to improve the applicability of the system. The robust target detection problem in a complex target background is vital to improve the applicability, thus, the key problems are: 1) statistical modeling of complex backgrounds; 2) The robustness of the detection technology.In views of this, the thesis is concerned with the target detection problem of moving objects in a complex background, and proposes some detecting algorithms, specifically, the main work in this thesis can be summarized as:Firstly, after studying the principles and analyzing the advantages and shortcomings of traditional inter-frame difference method. We choose the frame difference sequence of each pixel as the object model, useing Gaussian mixture model for data modeling.Secondly, we made a study of Statistical Theory of Target Detection and applied concepts of false alarm rate detection. Based on method of threshold detecting, the function between threshold and false alarm rate is established. Then a linear regression-based strategy for background updating is proposed. After Differential images are binarily processed by thresholds computed through the above method, the contour of moving object is clearly visible, around which the complex background is transformed into discrete punctate noise.Finally, after comparing various filtering and edge detection methods, we proposed a de-noise and foreground segmentation method which is base on finding geometric center of non-zero pixels.A large number of simulations are provided to illustrate the effectiveness of the proposed theories.