Moving Objects Detection Based On Real-time Video

Abstract：Moving objects detection is a technology extracted interest targetfrom a video image sequence, is other video analysis technology prerequisiteof target classification, target tracking, target behavior, and also is focus ofmachine vision and digital image processing research. So, research onalgorithms for moving objects detection in video sequences has importantresearch value and significance.Firstly, this paper briefly introduces the status quo of the moving objectsdetection algorithms at home and abroad, then, analyses classic movingobjects detection domestic and oversea, lastly, carry out a detailed analysis ofthe advantages and disadvantages of the moving objects detection algorithms.For the validation of effectiveness of the proposed algorithm, this paperselects several hot algorithms include frame difference method, Gaussianmixture background modeling algorithm and Vibe algorithm for comparisonand analysis.We find that Vibe background modeling has advantages of consumefewer resources and fast speed after study on it for long period of time, so, itis very applied to real-time moving objects detection. However, there is thefollowing questions such as stationary target problems, ghosting problem,shadow eliminate problems. Vibe algorithm for these problems givesimprovements. For stationary targets, adding prospects judgment module toprevent moving target which will remain for a long time to absorb intobackground. For ghosting problem, adding histogram matching to dodiscriminant, thereby can effectively eliminate the ghost. For shadowelimination problems, making a shadow detector model according to thedifference of background and shadow, it can eliminate the target shadow insome extent. Although the effect of the Vibe algorithm in moving objectsdetection is improved, it still appears the incomplete phenomenon of thetarget area. Therefore, this paper proposes a novel moving target detectionalgorithm consisted of improved Vibe and matting techniques in image processing. The algorithm quickly and accurately locks the prospects of thetarget area under the premise of the improved Vibe algorithm, then, takingadvantages of incomplete label of matting technology and high-precisionextracting the full movement of the target area. So, this algorithm is a goodsolution to the movement target detection area of empty, edge incomplete likethe shape of “c” font and fault issues.This paper combines the improved Vibe algorithm and mattingtechnology mainly including Grabcut and closed form matting, then, Verifythe accuracy, real-time and robustness of the proposed algorithms by meansof comparative analysis of these two combined strategy.