Abstract: Varying illumination, noise of camera and inadequate background update are the challenges in the current video object detection. A moving object detection method with tracking compensation in color images based on spacetime background difference was proposed. First, this paper established a Gauss background mixture model; and eliminated the varying illumination by adding impact factor in the color image difference. Then frame difference and domain difference were fused with moving tracking seeds compensation to get the seeds of real moving objects; based on the rough foreground which was got by background difference, according to the filter method based on the moving seeds which connected the region of moving objects, the final real and integral foreground target was found. The data collected from many real surveillance videos show that the accuracy and integrity of object detection are greatly improved by using this method.

Key words: space-time background difference, varying illumination, tracking compensation, motion detection

摘要： 视频目标检测中，光照变化、摄像机噪声和错误背景更新是目前的难题。提出一种带跟踪补偿的时空背景差彩色图像运动目标检测方法。首先，建立混合高斯背景模型，通过在彩色图像差分中加入影响因子消除光照变化；接着，通过帧间差分及邻域差分，加上运动跟踪种子补偿，获得真实运动目标的种子点；在背景差分粗前景基础上，根据连通区域运动种子点过滤法，检测出最终的真实而且完整的前景目标。通过多个实际视频监控的视频数据集的实验，结果表明该方法的目标检测准确率和完整性均有很大程度的提高。

关键词: 时空背景差, 光照变化, 跟踪补偿, 运动检测