背景渐变的视频对象分割算法研究及实现

提出了一种有效的背景渐变的视频对象分割算法.首先将前一帧分成前景和背景两部分,然后采用灰度投影匹配算法对当前帧进行全局运动估计和补偿,将当前帧与上一帧进行差分运算,便可得到差分图像.通过对差分图像进行二值化处理,得到运动模板并与前景信息进行相与计算,再结合当前帧信息便可得到运动目标.在TI公司的TMS320DM642芯片上验证了该算法,实验结果表明该算法不仅对亮度变化和环境变化具有鲁棒性,而且可独立、精确地分割出运动目标.

Segmentation of Moving Object in Video Sequence Having Gradual Changing Background and Its DSP Implement

This paper puts forward an effective moving object segmentation algorithm on moving video sequences. First, the first frame is divided into the foreground and background. Then,global motion compensation is made on the current frame after global motion estimation, and differential operation is performed between the current frame and the previous frame which are filtered in advance. Then, by combining the motion region and the foreground, the video objects can be generated. We use the TMS320DM642 to test and verify our algorithm. Experimental results show that the proposed algorithm is effective and very robust.