Targets Detection Based On Mixture Gaussian Model And Direction Detection Of Optical-flow

The detection of moving object is an important area of machine vision research,widely used in intelligent security devices. There are many major areas of machinevision such as image processing, pattern recognition and motion detection. MovingObject Detection can be used in a variety of surveillance cameras, or industrial robots.Background subtraction, optical flow method and frame difference can be used tovideo sequence to detect the moving object, premise of object classification andtracking. Therefore, due to the different physical environment, such as indoor andoutdoor, different scene changes and scene of different factors, different detectionmethod have its limitations. According to the scope of the difficulties of the existenceof a variety of detection methods and their advantages and disadvantages choose asuitable detection method can reduce the computational complexity can be a good testresult. Experiment object of this article is fixed scenes indoor, the request is to detecta complete moving object and position information of subsequent operations, sochoose the background subtraction based on Gaussian mixture background modeling,the research work mainly conducts in-depth study of its background modeling andobject detection methods, target detection incomplete caused by slow movement,initialization of background modeling parameters also has made some improvements,determine the direction of moving object. The main contributions are as follows:1.we introduced the principle and algorithm of some moving target detectionmethods, especially to the Gaussian mixture background modeling detectionalgorithm. Comprise the assignment and update of some initial parameters and themodel of Gaussian distribution, method to background modeling and extraction themoving foreground. Improved method of the incomplete detect caused by slowmotion.2. Introduces some basic image processing method, a series of processing ofthe video sequence before the experiment, some necessary erosion and dilationoperations to the foreground image effectively removes some noise disturbanceformed for follow the direction of motion detection accuracy has laid a solidfoundation.3.In direction detection module introduce the principle of optical flow algorithm,set thresholds to the moving target while detected by background subtraction, select moving target which meet the requirements and set a region of ROI, find the featurepoints from it and find the same feature points from the next frame by use pyramidoptical flow. Set the light flow direction as the moving direction of the target area, thetarget area and set conditions only in the vertical direction of motion detection.