项目名称: 基于时空间排他性块匹配法的人物行为识别技术研究
项目编号: No.61302134
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 无线电电子学、电信技术
项目作者: 李竹
作者单位: 杭州电子科技大学
项目金额: 27万元
中文摘要: 基于计算机视觉的个体人物行为的自动识别,可以广泛应用于民生、公共安全、工商业活动等各个领域。传统的识别技术往往分为通过提取人物的运动轨迹,描述轨迹特征和特征匹配几个步骤。而运动轨迹提取的精确度直接影响识别的正确率。本课题将一种申请者自主开发的新的轨迹提取算法,即排他性块匹配法(exclusive block matching),应用于行为识别中。与传统的方法相比排他性块匹配法可以更为完整的提取轨迹,同时提取的轨迹又具有更高的的精确度和鲁棒性,可以有效的提高行为识别的正确率。此外,本课题还提出一种新的行为轨迹的描述方法,即轨迹上下文直方图(Context Histogram of Trajectory)。这种方法可以描述极为复杂的运动轨迹,因此可以有效地发挥排他性块匹配法的优势。本课题的研究成果除了可以直接应用在监控类产品中以外,也可以为未来的群体行为识别技术研究打下坚实的基础。
中文关键词: 人物行为识别;轨迹;块匹配;硬件加速;特征描述
英文摘要: Human activity recognition based on computer vision can be widly used in public security,livehood, commerce and industry. Traditional human activity recognition technique includes three steps which are extracting trajectories, describing trajectories and matching. In this study, we apply a new technique of feature extraction and matching to activity recognition. This method which is developed by ourself in previous study is called Exclusive Block Matching(EBM) method. Compared with traditional method, it can extract trajectories of all parts of humans with higher robustness and accuracy rates. Moreover we also propose a new method to describe trajectories. This method is called Context Histogram of Trajectory(CHOT) method. It can describe more complex trajectories than traditional methods. This advantage of CHOT method makes the EBM method can give fully play when both of them are applied in activity recognition. The new technique proposed in this study can be used in monitoring system derectly. Moreover,it provides research foundation for our future works of group activity recognition.
英文关键词: Human activity recognition;trajectories;block matching;gpu;feature description