中文摘要：

提出人体原子行为的概念，通过将人体做为一个质点，研究其运动轨迹的特征进行在线识别。对得到的轨迹进行归一化处理，去掉拍摄距离、拍摄角度和行走方向的影响，按照原子行为的周期特征对轨迹进行分割；在产生轨迹周期期间，不断更新行为轨迹的运动特征；在原子行为周期结束前，采用K近邻法对行为进行识别。实验结果表明，该技术对环境光照的要求低，实时性强，识别率高，适用于单人或多人的在线行为识别系统。

英文摘要：

Human atom action conception was proposed Regarding a human body as a particle,human action formed a trajecto-ry.On-line action recognition was realized by studying characters of the trajectory.The trajectory was normalized to get rid of the influences from photographic distances,angles and the human movement directions.The trajectory was segmented according to characters of human atom action.Values of character parameter were updated from action trajectory during the action period.The K nearest neighbor algorithm was used for action recognition before the end of an atom action period.Experimental results show that this technology,which is robust to the ambient illumination,has a high recognition rate with high real time capability.It is suitable for single or many people on-line action recognition system.