标题:
基于Kinect的人体危险动作检测Human Dangerous Action Detection Based on Kinect
作者:
李晓林, 吕周南, 孙凤池
关键字:
危险动作, 骨骼跟踪, 跌倒检测, 模式分类, 支持向量机Dangerous Action, Skeletal Tracking, Fall Detection, Pattern Classification, Support Vector Machine
期刊名称:
《Journal of Sensor Technology and Application》, Vol.4 No.1, 2016-01-22
摘要:
针对家庭环境中老年人的健康安全问题，提出了一种基于Kinect骨骼数据的人体危险动作检测方法。对日常生活中的人体动作模式进行分析，将对人体直接造成伤害和预示着人体即将受到伤害的两种动作定义为危险动作。利用Kinect传感器提供的骨骼跟踪获取人体的头部位置，对不同动作模式下人体头部位置的变化规律进行分析。根据头部位置变化作为危险动作检测的特征，通过支持向量机分类器对人体的动作模式进行分类，可以有效检测出日常家居环境中的危险动作，与基于头部运动速度的方法相比，误判、漏判现象明显减少，识别正确率较高，且具有良好的可扩展性。
In view of the health and safety of the elderly in the family environment, a human dangerous action detection method based on Kinect skeletal data is proposed. The action that will directly damage human body or presage the happening of dangerous situation is defined as dangerous action by the analysis of human behavior in daily life. The head position of human body is obtained by using Kinect sensor and the change of head position in different action modes is analyzed. The action mode of human body is classified by support vector machine classifier according to the changes of head position as hazardous motion detection features so that the dangerous action in daily home environment can be effectively detected. Compared with the method based on the velocity of the head, misjudgment and false negative phenomenon are significantly reduced, and this method has good scalability and high recognition accuracy.