一种基于分层的人类动作识别和定位算法研究

Research on A Human Action Recognition and Positioning Algorithm Based on Hierarchical

摘要:

人类动作识别在视频自动分析、视频检索等领域获得广泛应用,是目前的研究热点。然而现有的动作识别方法重点关注视频的非静态部分而忽略大部分静态部分,从而影响了动作识别和定位的效果。本文提出一种新的分层空间-时间分段表示法,以分层方式实现部位和整个身体的多分辨率表示,可用于运动识别和定位。该算法分为3个步骤。第一步,首先对每个视频帧进行分层分段,以得到一组分段树,每颗树是身体分段树的候选。第二步,利用视频的轮廓、接合对象结构、全局前景色等信息对候选分段树进行修剪。第三步,在时域上对剩余分段层的每个分段进行前向和后向跟踪。我们以难度较大的UCF-Sports和HighFive数据集为实验对象,对本文方法进行性能评估,实验结果表明,本文方法的性能要优于当前最新运动检测算法性能,运动定位性能与当前最新算法相当。

Abstract:

Human action recognition is a hot topic of research, due to its wide ranging application in automatic video analysis, video retrieval and more. However, the existing action recognition methods focus on non-static parts of the video, while the static parts are largely discarded. This is affecting the accuracy of action recognition and location. In this paper, a new hierarchical space-time segments representation designed for both action recognition and localization that incorporates multi-grained representation of the parts and the whole body in a hierarchical way. The proposed algorithm comprises three major steps. We first apply hierarchical segmentation on each video frame to get a set of segment trees, each of which is considered as a candidate segment tree of the human body. In the second step, we prune the candidates by exploring several cues such as shape, articulated objects’ structure and global foreground color. Finally, we track each segment of the remaining segment trees in time both forward and backward. The experimental results show that, the performance of our method is better than the state-of-art action recognition methods on two challenging benchmark datasets UCF-Sports and HighFive, and at the same time produce good action localization results.