一种视频对象提取与跟踪的新方法

针对视频对象的分割问题，提出了一种新的视频对象提取与跟踪方法．首先采用分水岭分割和区域合并的方法进行初始分割，再通过光流场估计和全局运动估计计算全局运动的残余误差，在分割的每个区域上利用残余误差数据进行假设检验以确定运动区域，组合所有的运动区域即可提取完整的视频对象．然后使用双向投影的方法在后续帧中快速准确地跟踪视频对象，前向投影用来定位视频对象和简化分割计算，后向投影则用来确定每个分割的区域是否属于跟踪的视频对象．对MPEG-4测试序列的实验结果表明，本方法具有良好的分割性能．

A New Approach to Video Object Extraction and Tracking

For the problem of video object segmentation, a new approach was presented to video object extraction and tracking. The watershed segmentation and region merging are used to obtain an initial partition. The global motion residual error is calculated with the results of optical flow estimation and global motion estimation. Moving regions are identified by hypothesis testing on the residual error data of each segmented region. The whole video objects are extracted by grouping all moving regions. Then, a bidirectional projection scheme is proposed to efficiently track the extracted video objects in the subsequent frames. Forward projection is used to localize the video objects and reduce the computation cost of segmentation, and backward projection is employed to decide whether each segmented region belongs to the tracked video objects or not. The experimental results for the MPEG-4 test sequences demonstrate the good segmentation performance of the proposed approach.