在康复理疗系统中基于人体部位的多人检测跟踪方法

MULTIPLE TARGET DETECTION AND TRACKING METHOD BASED ON HUMAN BODY PARTS IN THE REHABILITATION PHYSIOTHERAPY SYSTEM

针对在当前康复理疗系统中人体跟踪存在的问题,需要实时观测被监视场景中的运动目标,分析描述他们的行为,通过对人体行为的理解和判断,才能得到相应的结论并做出相应的决策。提出一种基于人体部位的支持向量机分类器的方法,实现人体跟踪。这种方法能够捕捉在人体姿态和背景变化时的人体关节部位,利用训练的人体部位模型能够在人相互遮挡时正确检测人体部位。在检测阶段,选择一个人体部位子集,最大限度地提高检测概率,大大提高了在多人场景中的检测性能。在跟踪阶段,利用SVM分类器实现人体的有效跟踪。实验表明该方法能够在人体相互遮挡情况下,正确检测和跟踪人体。

In view of the problems that exist in human body tracking in the present rehabilitation physiotherapy system,it needs real-time observation of the moving target in the scene under monitoring,analyze and describe their behaviors. Only by understanding and judgment of human body behaviors can people obtain corresponding conclusions and make corresponding decisions. The paper proposes an SVM classifier method based on human body parts to realize human body tracking. The method can capture human body joints when human posture or background changes. By referring to the trained human body model it can correctly detect human body parts even when they block each other. In the detection stage,it selects a human body part subset to ultimately improve the detection probability,so that it greatly enhances its detection performance in scenes where there are multiple people. In the tracking stage,it utilizes SVM classifier to realize effective human body tracking. Experiment show the method can correctly detect and track human bodies when they block each other.