视频序列中的运动目标检测算法研究

Research on moving target detection algorithm in video sequence

视频序列中前景目标的提取存在因光线发生突变而漏检前景目标,以及背景包含动态信息时识别错误等问题。由于光线突变会产生大面积的阴影,并且包含动态信息的背景会产生干扰和噪声,提出了一种Vibe算法的改进应用,Vibe能够快速有效地抑制阴影,并且结合了帧间差分法对突变不敏感性的特点,使检测前景目标更加准确。分别针对光线突变以及背景包含动态信息的2种情况进行实验验证,同时用光流法、GMM合背景建模、codobook算法、传统Vibe算法与改进的Vibe算法分别对2种情况视频序列进行处理,之后对各个算法进行定量分析,提取结果和定量分析结果是相匹配的。实验结果表明,改进的Vibe方法可以使在前景目标在光线突变或背景包含动态信息的情况下,不被漏检并且可以更好地去噪,提取更加完整的前景目标。

The extraction of foreground object in video sequence has problems such as missed detection of foreground target due to sudden changes in light,and recognition errors when the background contains dynamic information.A sudden chagen in light will produce large shadow,and the background contains dynamic information that will cause interference and noise,the paper gives an application of improved Vibe algorithm.Vibe can effectively suppress the shadow,and the combining with the insensitivity of inter frame difference method to light mutation the accuracy of foreground target extraction is improved.Experiments were carried out to verify the light mutation and the dynamic information of the background.At the same time,the optical flow method,Gaussian mixed background modeling,clustering codebook algorithm,traditional Vibe algorithm and improved Vibe algorithm are applied to the two cases of video sequences.After the quantitative analysis of each algorithm,the extraction results and the quantitative analysis results are matched.The experimental results show that this method can make the foreground target not be missed,denoise in the case of sudden change of light and background containing dynamic information,and can extract more complete foreground target.