The Research On Moving Object Detection Algorithm In Video

With the rapid development of video analysis technology, the moving object detection technology has very wide practical application value. Today, not only in the densely populated and larger flow schools, airports, residential district, large shopping malls, and the daily other areas, but also in areas, such as traffic monitoring, military security, medical cells also pay close attention to by scholars. Now, scholars have proposed many moving object detection algorithms, however, various algorithms have some limitations, though some algorithms can detect complete moving object, but they are strongly influenced by illumination mutation and noise, some algorithms can eliminate the above problems to some extent, but the large amount of calculation, processing speed is slow, and cannot achieve the real-time demand of object detection. Therefore, looking for a kind of high accuracy, processing speed and good real-time moving object detection algorithm is an important challenge that we are facing.This article main research content is as follows:First of all, the basic theoretical knowledge is studied in digital image processing technology, such as graying, image filtering, thresholding, feature extraction, morphology processing, connected domain analysis and Otsu. And studied three kinds of classic methods of object detection: optical flow, background subtraction, inter-frame difference, in view of the background subtraction, and introduces five kinds of common background model methods. At last, this paper selects the background subtraction and three-frame difference method to establish model of moving object based on Gaussian mixture modeling division by comparison.Second, because of the background of the traditional Gaussian mixture model initial established, and noise robustness is not well, it is easy to detect errors, this article first proposes a Gaussian mixture model of adaptive update rate, then background model using LBP texture feature updates again to get more realistic background. Three-frame difference method for existing cavity in the object detection and object edge detection has incomplete problem, to put forward six-frame difference and Sobel edge feature fusion improved methods.Finally, because two improved algorithms have complementarity for adaptive of complex scenarios and sensitivity of light change cancellation, therefore, two kinds of joint improved algorithm builds an efficient moving object model. At last, through the connectivity testing and morphological processing, to get the complete exercise foreground pixels. Object detection results show that different scenarios, which solves effectively the problems when the illuminance abrupt variation, noise interference, light cavity and double phenomenon, and the improved algorithm has more robustness that compared with the same kind of other algorithms.