关键词: 运动目标检测, 贝叶斯决策, 颜色同现, 颜色相似, 模糊点

Abstract: For the purpose to improve the moving objects detection performance of the traditional Gaussian background modeling 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 accurately.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.

Key words: moving object detection, Bayesian decision, color co-occurrence, color similarity, confusion point