基于区域相关性的全局运动视频中非刚性对象跟踪

Tracking Non-rigid Object Based on Region Relativity in Global Active Video

提出一种适合全局运动视频中自动探测与跟踪非刚性对象的OT-GAV模型.该模型首先利用基于区域相关性的RDM算法计算相邻帧区域匹配,并结合Q学习与K-S统计法优化匹配结果,获得较为精确的区域运动向量.然后,利用前景和背景存在的运动形态差异,区域动态纹理一致性及对象运动过程中保持区域完整性的特点,逐步实现前景对象区域的探测与合并.实验证明,本模型及其相关算法可在室内和室外环境下,自动探测前景关注对象,获得其较为精确的边缘信息,并实施有效的跟踪.同时,该模型还能够解决对象跟踪过程中的"空洞"问题.

Presents an OT-GAV model for automatic detecting and tracking non-rigid objects in global active video. The model computes region matching between adjoining frames based on RDM algorithm. Q-learning and K-S statistic are used to optimize matching results so that the region motion vectors are precise in a certain extent. Then motion modality difference between foreground and background, consistency of region dynamic texture, and region integrality during object moving are gradually utilized to realize object regions detecting and combination. The experiment proves this model and correlative algorithms could automatically detect foreground attentive objects, obtain precise edge information and effectively track in outdoor or indoor circumstance. Besides, this model is able to resolve empty-hole problem during object tracking.