论文部分内容阅读
在不同的光照及视角下,为了实现人体日常生活动作的高识别率,提出了一种基于Kinect的识别方法.首先,受到人类进行动作识别时往往关注局部细节动作的启发,层次化地处理了采集到的人体关节点数据:通过判断躯干关节点位置变化的缓慢程度,将动作粗分类为上肢动作和躯干动作;之后对于上肢动作,关注手部关节轨迹变化,而对于躯干动作,关注中心关节点轨迹.然后,通过C均值聚类法从这两类轨迹中提取关键点,并将动作的轨迹映射到相应的关键点,得到每组粗分类动作的关键点序列.并提出了时序直方图的概念用以建模关键点序列.再通过比较轨迹间关键点序列的相似性,完成动作识别任务.最后,将该算法应用于采集的数据集合,得到了99%的识别正确率,表明算法能够有效地完成人体动作识别任务.
In order to achieve a high recognition rate of human daily life movements under different light and visual angles, a Kinect-based recognition method is proposed.Firstly, when human beings recognize the motion, they often focus on the action of local details and hierarchically handle Collected human joint point data: by judging the degree of trunk joint position changes slowly, the action is roughly classified as upper limb movements and torso movements; followed by upper limb movements, hand joints trajectory changes, while the trunk movements, focus on the center joint Then the key points are extracted from these two types of trajectories by C-means clustering method, and the trajectories of the actions are mapped to the corresponding key points, and the key points of each group of rough classification actions are obtained, and the timing histogram Is used to model the sequence of key points.And then the motion recognition task is completed by comparing the similarity of the key points between trajectories.Finally, the algorithm is applied to the collected data sets, and the recognition accuracy of 99% is obtained, which shows that the algorithm Can effectively complete the task of human motion recognition.