中文摘要：

针对复杂场景中包含的摄像机扫描运动、随机抖动和目标运动,提出一种基于帧间可变块差分的运动目标检测算法。首先,利用全局特征点估计运动参数对帧间背景进行补偿,提取图像的全局特征点并匹配,以特征点集的最小位置误差和作为目标进行迭代,获取误差不大于0.5 pixel的全局运动参数,并精确补偿当前帧实现背景校正。然后,利用可变块差分实现运动目标的检测。先用大尺寸对差分图像进行分割,将整幅图像粗略区分为背景区域、运动目标区域和边界区域,通过阈值判定来进一步细分,最后对运动目标区域进行准确标定。这种由粗到细的检测步骤,在保证精度的同时能够提高检测速度。实验结果表明,该算法能够准确检测含摄像机扫描和抖动的复杂运动场景中的前景运动目标,且处理速度达到25 frame/s。

英文摘要：

A fast moving-object detection algorithm based on inter-frame Variant Compensated Blocks Difference（VCBD） is presented to deal with the complex scenes with camera scan,dithering and object moving.Firstly,the global motion estimation is performed based on feature points to compensate the inter-frame background.The global feature points in a reference image are selected and matched in a current image.Then,the iteration is applied to realize the minimum sum of position errors of all matched points and to obtain the global motion parameter with the accuracy less than 0.5 pixel.Accordingly,the current frame is compensated to match the background area.Finally,the adaptive variant block difference is proposed to detect moving objects.The whole image is classified into background,foreground and boundary areas and the block is then judged with the threshold and divided into four blocks.These coarse-to-fine steps can greatly improve the velocity and veracity of detection.Experimental results show that the algorithm can detect moving objects in camera scan and dithering sequences and the processing speed achieves 25 frame/s.