Abstract: Aiming at the activities directions of human are arbitrary, this paper proposes the viewpoint independent behavior recognition method. A camera with fixed position is used to measure the motion direction based on the change of mass center coordinates in world space. According to the direction, video sequences can be selected taken by the camera which visual angle is orthogonal or parallel to the movement direction. A database on key features of motion activity sequences is taken orthogonally and parallel to the subject’s movement is established for several human behaviors indoors. The method of parallel line bound with single camera is used to reconstruct 3D coordinate of key points by image coordinate and conversion factor. The method is tested on sequences of video data of human motions indoors, and experimental results validate the good performance of the proposed approach in viewpoint independent behavior recognition.

Key words: behavior recognition, view independent, motion direction, intelligent video supervision

摘要： 针对人体运动方向的随机特性，研究场景中有多台摄像机时视角无关的行为识别方法。利用一台位置固定的摄像机，根据视频序列中运动目标质心空间坐标变化，确定其运动大致方向。根据该方向，选择具有垂直和平行运动轨迹方向视角的相机获得的序列图像，进行行为分析。提取人体运动行为侧像和正像轮廓的关键姿态建立特征库。应用单个相机平行线约束，通过转换因子由图像坐标恢复运动目标关键点的三维坐标，从而确定目标运动方向。建立室内多视角行为视频样本库，实验结果表明，利用该方法识别室内运动行为，能够达到视角无关行为识别的目的。

关键词: 行为识别, 视角无关, 运动方向, 智能视频监控

CLC Number: 

• TP391.41