光学精密工程
光學精密工程
광학정밀공정
OPTICS AND PRECISION ENGINEERING
2009年
6期
1485-1495
,共11页
机器学习%片图模型%改进的隐马氏模型
機器學習%片圖模型%改進的隱馬氏模型
궤기학습%편도모형%개진적은마씨모형
machine learning%pictorial structure model%modified Hidden Markor Model(HMM)
基于人体模型的跟踪方法易于实现视频的运动人体跟踪,而且利用较少的视频帧数即可学习获得人体模型.本文针对给出的视频提出了学习人体模型的学习算法.利用片图模型表示未经学习的人体,改进的隐马尔可夫模型(HMM)模拟人体在视频序列各帧间的运动,并使用机器学习方法对该改进的HMM进行推理,获取改进HMM的参数,从而获得所需的人体模型.学习得到的人体模型由包含颜色信息的各人体肢体模板组成.实验显示只用80~90帧包含有人体运动的序列图像,便可学习得到该运动人体的人体模型.结果表明,该学习框架效果明显,可用于快速学习视频序列中的运动人体模型,且可用于学习一人或多人的人体模型.
基于人體模型的跟蹤方法易于實現視頻的運動人體跟蹤,而且利用較少的視頻幀數即可學習穫得人體模型.本文針對給齣的視頻提齣瞭學習人體模型的學習算法.利用片圖模型錶示未經學習的人體,改進的隱馬爾可伕模型(HMM)模擬人體在視頻序列各幀間的運動,併使用機器學習方法對該改進的HMM進行推理,穫取改進HMM的參數,從而穫得所需的人體模型.學習得到的人體模型由包含顏色信息的各人體肢體模闆組成.實驗顯示隻用80~90幀包含有人體運動的序列圖像,便可學習得到該運動人體的人體模型.結果錶明,該學習框架效果明顯,可用于快速學習視頻序列中的運動人體模型,且可用于學習一人或多人的人體模型.
기우인체모형적근종방법역우실현시빈적운동인체근종,이차이용교소적시빈정수즉가학습획득인체모형.본문침대급출적시빈제출료학습인체모형적학습산법.이용편도모형표시미경학습적인체,개진적은마이가부모형(HMM)모의인체재시빈서렬각정간적운동,병사용궤기학습방법대해개진적HMM진행추리,획취개진HMM적삼수,종이획득소수적인체모형.학습득도적인체모형유포함안색신식적각인체지체모판조성.실험현시지용80~90정포함유인체운동적서렬도상,편가학습득도해운동인체적인체모형.결과표명,해학습광가효과명현,가용우쾌속학습시빈서렬중적운동인체모형,차가용우학습일인혹다인적인체모형.
The tracking method based on a people model contributes to realizing the kinetic people tracking for a given video,and it can learn the people model by using much less frames of the video. This paper proposes a learn algorithm for learning the people model in the given video.By using a tree pictorial structure model to represent the detuned generic people in the video, and a modified Hidden Markov Model(HMM) to simulate the motion of people between the two frames of the video, a machine learning method is used to the modified HMM to obtain the estimation of parameter of the modified HMM,and to capture the people model from the video. The learned model consists of different body templates covered with color information. For learning the color of the local parts of the people model by proposed algorithm,an instance-specific model has been obtained.The experiment demonstrates that the kinetic people by proposed algorithm model can be learned with sequences images of 80-90 frames involving people motion,which shows the learning method works well for learning the kinetic people model based on video, and can rapidly length models for one or more persons in the video.