The Research Of Key Technologies On Moving Targets Detection Under Complicated Scenes

As the rapid development of the computer and the increasing popularity of imageacquisition equipment, moving targets of complicated monitoring scenes in the videos are theimportant research objects. Compared with the manual way, detecting moving targetsreal-time, using digital image processing and pattern recognition technology can realizeautomatic and fast detection to moving targets in complicated scenes, which not only decreasethe workload of administrators, and also improves the overall performance of moving targetsdetection in monitoring scenes. At present, moving targets detection based on digital imageprocessing and pattern recognition has been widely used in the transformer substations, banks,sensitive military area, shopping mall and battlefield, and so on, furthermore, which becomethe focus of researching.Aiming at the complicated monitoring scenes whose key monitoring area and non-keymonitoring area have significant degrees of distinction, this paper will focus on three keyparts for study: accurate extraction of key monitoring regions, fast motion targets detectionalgorithm based on background substation method and motion targets detection usingdifference of adjacent frames based on edge image, in order to propose a rapid and robustmotion targets detection algorithm combining background difference with edge framedifference.In aspect of extracting key monitoring regions, this paper analyzes the characteristics ofkey monitoring regions in different typical complicated scenes, researches on the existingmethods of the key monitoring regions. Then, a new algorithm of key monitoring regions extraction was proposed according to the background color statistic model. First of all, westudy the different channel in color space effect on degrees of distinctions between keymonitoring areas and non-key monitoring areas. Secondly, select the V channel of HSV colorspace, according to the color distribution of the channel V in key monitoring areas, modelbuilding of sampling color using a single Gaussian function. Finally, based on the establishedmodel of the V channel, study on the key monitoring regions’ segmentation and extraction,use morphological processing to accomplish precise recognition of the key monitoringregions.Based on the background difference, we aim to put forward new thinking that can meetthe real-time moving targets detection and quickly track background dynamic changes.Median filtering can characterize the real background better if the foreground is Gaussiandistribution. So this paper uses median filtering algorithm for initial background modeling.On this basis, in the process of moving object detection, establish background count matrix ofkey monitoring regions, use the different credible degrees of the pixels in the matrix to updateeach pixel’s background values with different updated rate. Compared with moving targetsdetection using the background differential method by single fixed update rate, our methodcan track the changes of the background more quickly, and then improve the moving targetsdetection rate.The way based on difference of adjacent frames has good real-time performance and theimage edge has good resistance to the noise, based on edge difference of adjacent frames canreflect the moving targets’ detail outline. As a consequent, this paper combines fastbackground difference method with edge difference of adjacent frames method. It not onlyimproves the integrity of moving target detection, but also improves the accuracy to detect themoving targets.Finally, experimental study has been conducted by using the sequence images undertypical complex monitoring environment. The results show that the algorithm proposed in thispaper can effectively achieves the key monitoring regions’ segmentation, the key monitoringregions’ background modeling, moving targets detection accurately. All of these promote thedevelopment of intelligence monitoring for moving targets in complex monitoringenvironments.