Font Size: a A A

Moving Object Detection Research Based On Video

Posted on:2015-06-30Degree:DoctorType:Dissertation
Country:ChinaCandidate:M HuangFull Text:PDF
GTID:1108330491963064Subject:Image Processing and Scientific Visualization
Abstract/Summary:PDF Full Text Request
Video analysis is one of the essential fields of computer vision and pattern recognition, the core technology of which is the detection, recognition and tracking of the object in the field of view. The detection of moving object is the prerequisite and foundation of video analysis. The accuracy, completeness and efficiency of subsequent recognition and tracking rely heavily on the precise detection of the shape and location of the target object. Consequently, the research of moving object detection boasts important value in academic study and application.Numerous methods for moving object detection have been reported in the literature. The main methods adopted by researchers include optical flow, background differential, inter-frame differential, wavelet method, extended EM algorithm, scene change detection based on mathematical morphology and the combination of various algorithms. These basic algorithms reach satisfactory results since they take different perspectives and provide solutions to moving object detection based on various situations; however, there are some questions to be further studied due to the complex detection environment, such as background problem, shadow problem, block issue, etc.To research and resolve the above mentioned problems, many researchers focus on background differential method, the key of which is to effectively extract and update objects from the background. The main approaches taken are:statistic average background model, frame differential background model, single-mode Gaussian background model and Gaussian mixture background model. The progress of these methods promotes the accuracy of moving object detection. Inspired by the previous researches, we have studied the shot detection and key frame extraction in the video sequence. Under this prerequisite, we discuss the background update strategy based on key frame and the detection of moving objects in the video by the combined algorithm of differential matrix of pixels and threshold determination. The experiment results reveal that noticeable better detection effect can be reached. We also propose an algorithm based on background models of HSV Color Gaussian and Key Frame HSV Color Gaussian to conduct moving object detection with connectivity principles and texture gradient. The detection result avoids the "Hole" phenomenon during the process and thus acquires the contour of the moving object with more precision; meanwhile, the shadow is removed effectively.Considering the characteristics of infrared image, i.e., low contrast, low brightness and small target sensitivity, we propose one target detection algorithm based on infrared image enhancement and one detection and extraction algorithm based on pixel probability of infrared image. They recognize the moving object in the infrared video through mixed cumulative histogram and color probability of pixel respectively.It is difficult to build target motion model because of the complicated actual situation and the unpredictability of the moving mode. In order to resolve the above problem, the feasibility of using differential evolutionary algorithm for moving object detection based on MRF model is discussed. Finally, a moving object detection method based on Markov Random Field and distributed differential evolution algorithm is presented.This paper centers around the study of moving object detection and especially focuses on background updating algorithms, moving object detection based on statistics as well as their applications. The major study of this paper, specifically, is on the following aspects and reaps due achievements:(1) Research the extraction of key frame using segment video and apply this method to detect the movement in dynamic background; (2) Research the detection method based on threshold matrix of pixels, connectivity principles and texture gradient; (3) Research infrared image target detection based on infrared image enhancement and pixel probability; (4) Research MRF target detection method using special and temporal combination approach.
Keywords/Search Tags:object detection, key frame extraction, texture gradient, infrared video, MRF, differential evolution
PDF Full Text Request
Related items