Research Target Detection Method Based On Improved Gaussian Mixture Model

Gauss mixture model is an effective method in the field of image targetdetection. It can detect the abnormal changes in the environment timely, the basicidea of Gauss mixture model is that when environment match established model, it isbackground, otherwise is foreground. In this paper, we deep understudied andresearched current mainstream background model algorithm: the optical flow method,the inter-frame difference method, the background subtraction, meanwhile accordingto current domestic and foreign research progress of each algorithm, compared theadvantages and disadvantages carefully, At last selected Gauss mixture model in thebackground subtraction method as the basic method of this article. During targetdetection, Gaussian mixture model has some shortcomings in testing process, whenmodel replacing, we found that if not limited the process of replace, foreground iseasy to substituted into model mistakenly, impacting the result of detect. AfterGaussian mixture model checked, the result contains a lot of unusual circumstances,cause false detection in the system, and also exist problems that eliminate light isweak.for above situation, firstly we analyzed replacement process, and then accordingto the law of environmental change, eliminated interference factors to ensure thatnew background to replace outdated background; after Gaussian mixture modelchecking,we must further judgment to exclude all kinds of interference, keep actualmovement. Target detection, the prospect of judgment are pixels, information contentis often insufficient, cannot expression environment characteristic effective, in thispaper, it take the image in the independent change region judges as the unit, thenused the examination algorithm to distinguish disturbance and movement region;Finally, we made a judgment on the light conditions. Illumination occurs,background will change, we use algorithm to eliminate effect of light detection. Through this algorithm, the environmental testing achieved good results, andmaintained model stability, it can remove most of interference in foreground, andillumination solution has also made progress, but needs to further improve thedetection accuracy．...