面向家庭服务的人体动作识别

为了提高家庭服务机器人的智能,提出了一种面向家庭服务的人体动作识别算法。首先,提出了利用运动历史图像自适应更新背景的方法,对家庭环境进行背景建模;然后,利用YCbCr图像与灰度图像二值化的综合消除阴影,从而实现人体分割与定位;最后,提出了基于高熵变化量的时空特征方法,并结合环境信息对人体进行动作识别。实验结果表明,提出的背景模型能较好的克服暂时停止运动的物体被背景吸收的问题,并能有效地对背景的纯净度进行评价,同时,根据提出的动作识别机制,结合家庭环境的布置知识总体识别率可以达到95%。基本可以满足机器人家庭服务的要求。

Human Action Recognition Oriented to Family Service

In order to increase the intelligence of the family service robot, a human action recognition method is put forward. First, the scheme of Motion History Image is used to adaptively update the background model. Then, YCbCr images combined with gray images are used to segment the images to decrease the shadow. At last, space-time interest points based on high entropy changes are brought forward to recognize the human action. Experiments show that the background model can obtain the family environment background model satisfactorily. And the recognition rate of the action recognition scheme combined with the environmental information can reach 95%. It can satisfy the demand of robot family service.