摘要： 针对实际应用中待分割目标类型已知的情况，提出一种结合识别信息的多目标视频分割算法，使用训练数据集构建目标以及背景的特征字典，计算视频帧的超像素，构造一个分层条件随机场模型，用于约束视频帧的局部邻域和全局邻域，通过求解分层条件随机场模型，获得最终分割结果。实验结果表明，该算法能够对视频中相互遮挡及残缺不全的多个目标进行有效分割。

关键词: 多目标视频分割, 特征字典, 分层条件随机场

Abstract: The classes of objects are specific in most practical applications, an algorithm of multiple objects video segmentation with recognition information is proposed. The algorithm learns feature dictionary of object and background from training data, and constructs a hierarchical conditional random field model via computing super pixel for video frames, by which the local and global neighboring constraints in video frames are modeled. The final segmentation results are obtained by solving the hierarchical conditional random field model. Experimental results illustrate the algorithm can segment the objects of both occluded each other and the partial objects.

Key words: multiple objects video segmentation, feature dictionary, hierarchical conditional random fields

中图分类号: 

• N945