现代电子技术
現代電子技術
현대전자기술
MODERN ELECTRONICS TECHNIQUE
2015年
10期
55-60
,共6页
人体动作识别%极限学习机%协方差%方向位移直方图
人體動作識彆%極限學習機%協方差%方嚮位移直方圖
인체동작식별%겁한학습궤%협방차%방향위이직방도
human action recognition%extreme leaning machine%covariance%histogram of oriented displacement
基于关节信息的人体动作识别在人机交互、互动娱乐、多媒体信息检索等方面应用广泛。为了提高动作识别率，使用两种具有固定长度的分层描述符分别关注运动的动态和静态信息，对运动序列提取特征，将这两种描述符线性组合，形成同时包含动态和静态信息的新描述符，并使用极限学习机（ELM）进行分类。该方法在微软Kinect传感器采集到的MSRAction3D数据库和运动采集数据集HDM05上进行了仿真实验。实验结果证明组合后的描述符结合ELM在这两个数据集上的识别率较现有方法有明显提高。
基于關節信息的人體動作識彆在人機交互、互動娛樂、多媒體信息檢索等方麵應用廣汎。為瞭提高動作識彆率，使用兩種具有固定長度的分層描述符分彆關註運動的動態和靜態信息，對運動序列提取特徵，將這兩種描述符線性組閤，形成同時包含動態和靜態信息的新描述符，併使用極限學習機（ELM）進行分類。該方法在微軟Kinect傳感器採集到的MSRAction3D數據庫和運動採集數據集HDM05上進行瞭倣真實驗。實驗結果證明組閤後的描述符結閤ELM在這兩箇數據集上的識彆率較現有方法有明顯提高。
기우관절신식적인체동작식별재인궤교호、호동오악、다매체신식검색등방면응용엄범。위료제고동작식별솔，사용량충구유고정장도적분층묘술부분별관주운동적동태화정태신식，대운동서렬제취특정，장저량충묘술부선성조합，형성동시포함동태화정태신식적신묘술부，병사용겁한학습궤（ELM）진행분류。해방법재미연Kinect전감기채집도적MSRAction3D수거고화운동채집수거집HDM05상진행료방진실험。실험결과증명조합후적묘술부결합ELM재저량개수거집상적식별솔교현유방법유명현제고。
The human action recognition based on joint information has been widely used in human?robot/machine interac?tion,interactive entertainment,multimedia information retrieval and surveillance. In order to increase the action recognition rate,two hierarchic descriptors are adopted to extract the feature of the motion sequence,one of which focuses on the dynamic information and the other one focuses on static information. A new descriptor containing both dynamic and static information is created by lombining the two descriptors linearly. The extreme leaning machine(ELM)is chosen to make the classification. The simulation experiments for the method were carried out on the MSR?Action3D database and the HDM05 MoCap database. The results show that the lombined descriptor with ELM has significantly improvement in the accuracy of classification,both in the two databases,compared to the state?of?the?art methods.
