基于贝叶斯决策的运动目标检测方法

Moving objects detection based on Bayesian decision theory

传统混合高斯背景建模存在难以解决背景复杂以及阴影等因素影响视频运动目标检测效果的问题,为此提出了一种基于贝叶斯决策的运动目标检测方法。该方法利用帧间差分进行像素变化检测,将像素粗分为变化像素和非变化像素;对于变化像素中的运动点和静止点,通过统计确立有效的数据结构,其中显著颜色分布统计量用来描述静止点,而显著颜色同现统计量描述运动点;从数据结构中提取颜色特征矢量,将特征矢量中的静止点和运动点按照贝叶斯决策规则进一步分类为背景点、前景点和颜色相似点。对颜色相似点进行局部加权处理以达到正确检测的目的;通过融合静止点集、运动点集和加权后的颜色相似点集结果提取出前景运动目标。仿真实验表明,该方法能够在不同复杂背景下较准确地检测出视频中的运动目标,相比传统算法具有较强的鲁棒性。

For the purpose to improve the moving objects detection performance of the traditional Gaussian background mod- eling under the complex background, shadow, the method for moving objects detection based on Bayesian decision theory is proposed.The changed and unchanged pixels are classified using temporal difference.For the stationary and moving points among the changed pixels,effective data structure is established,where the statistics of most significant colors are used to describe the stationary parts of the background, and that of most significant color co-occurrences are used to describe the motion objects of the background.The stationary and moving pixels among the color feature vectors, which are extracted from the data structure, are further classified as background points, foreground points and the color similarity points by Bayesian decision theory.For the similar color pixels, local weighting is adopted to detect the moving objects accttrately.Foreground moving objects are extracted by combining the results of the stationary, motion and similar color pixels set.Experiments show that this method can accurately detect the moving objects under different complex background, and is robustness compared with traditional method.