基于语义轨迹的视频事件探测

Semantic Trajectory Based Video Event Detection

视频事件探测是视频内容自动理解领域的一个重要研究问题.在视频事件探测中,感兴趣对象的运动轨迹常被作为视频中探测事件的一种重要依据.目前基于轨迹的事件探测方法主要集中于根据轨迹几何特征进行视频事件探测,而忽略了与轨迹相关的语义信息.然而我们知道,轨迹的产生往往受到一些与轨迹相关联的语义信息的影响,如轨迹产生时的地理信息等.将轨迹相关联的语义信息整合到轨迹中可以使我们了解更多关于轨迹的信息.语义轨迹为我们提供了一个将语义信息与轨迹信息有效整合的方法.该文将语义轨迹应用到视频事件探测领域,提出了一个基于语义轨迹的视频事件探测方法.该方法将视频中抽取的感兴趣对象的原始轨迹转化为语义轨迹,并根据语义轨迹探测可能的视频事件.同时该方法还提供了一个描述语义轨迹特征以及对语义轨迹与轨迹特征进行匹配的方法.最后我们通过实验分析验证了基于语义轨迹的视频事件探测方法的有效性.

Video event detection is a key research problem for automatic video content under- standing. Trajectories of object are the one important clue for video event detection. Currently, the most trajectory based event detection methods focus on the geometric characteristics of the trajectory and neglect the semantic information related to the trajectory. However, as we known the trajectory producing process is affected by the trajectory related semantic information, such as geographic information related to trajectory etc. Combine the semantic information related to the trajectory with raw trajectory information can make us obtain more knowledge of trajectory. Se- mantic trajectory provides a way to effectively combine trajectory information with semantic knowledge. In this paper, the semantic trajectory is applied into video event detection and pro- posed a semantic trajectory based method for video event detection. This method can transform the trajectory of the interested object in video to semantic trajectory and detect the video event based on the semantic trajectory. Moreover, this method provides an approach for depicting se-mantic characters of semantic trajectory and matching the trajectory with trajectory characteristic description. Finally, the authors demonstrate the effectiveness of the proposed method through the empirical studies.