摘要
针对单台Kinect设备存在遮挡、传感数据丢失或错误等问题,文章引入了一种双Kinect测量3D人体关节信息的行为识别算法,首先结合Kinect下人体关节点的三维结构,利用两台kinect体感设备从不同视角采集人体行为数据,并通过两台Kinect设备空间坐标系的转化,为后续数据融合打下基础;同时,引入可信度参数、关节点变权重对采集的异常数据进行处理,并以改进卡尔曼滤波算法对缺失的关节点数据进行预测,以精准、完整的表征人体行为模型,进行人体行为识别。最后,通过实验分析,表明该算法在人体行为节点被遮挡状态下,仍能够通过数据融合方法,补足缺失的关节空间位置数据,实现对人体行为的全方位捕捉和识别。
Aiming at the problems of occlusion, loss of sensing data or errors in a single Kinect device, this paper introduced a behavior recognition algorithm for measuring 3D human joint information by double Kinect. Firstly, combied the three-dimensional structure of human joint points under Kinect, human behavior data were collected from different perspectives by two Kinect somatosensory devices, and the spatial coordinate system of two Kinect devices was transformed. At the same time, the reliability parameters and variable weights of joint points were introduced to process the abnormal data collected, and the missing data were predicted by improved Kalman filter algorithm to accurately and completely represent the human behavior model and recognize human behavior. Finally, the experimental analysis shows that the algorithm can still capture and recognize human behavior in an all-round way through data fusion method to supplement missing joint spatial location data when human behavior nodes are occluded.
作者
许荟蓉
Xu Huirong(College of Physical Education Yan’an University ,Yan’an 716000,China)
出处
《现代科学仪器》
2018年第6期48-52,共5页
Modern Scientific Instruments
