基于外形特征与运动特征的人体行为提取方法

针对传统外形特征表征方法描述行为动作能力有限和运动特征表征方法难以准确、稳定地捕捉目标运动特性等问题。提出运用人体外形特征和运动特征相结合的方法提取人体行为关键特征,利用谱聚类算法对特征进行降维,降低了数据维数,获得了最优的特征表征。仿真实验表明,该方法降低了样本维数,减少了数据冗余,并提高了训练精度,且保证了行为识别率。

Human Behaviour Extraction Method Based on Shape Features and Movement Features

According to the deficiency of the traditional shape feature representation method in limited ability to describe actions,and accuracy and stability problems of movement feature representation method in capture the target motion characteristics,this paper extracts human behavior key features by combining shape features and movement features. Spectral clustering algorithm is used to reduce the dimension of the features. Through reducing the dimension of the data,the optimal feature representation is obtained. Simulation experiments show that this method can reduce the dimension of the example and data redundancy,enhance training accuracy and recognition rate.