基于运动姿态描述子特征和词袋模型的行为识别

Behavior recognition based on motion posture descriptor and bag-of-words model

人体行为识别一直都是计算机视觉的研究热点,特别是随着微软Kinect传感器的出现,使得获取人体骨架信息更为便利。为了在骨架模型的基础上获得好的识别精度,基于三维的骨架序列信息,运用词袋模型和运动姿态描述子,结合改进的硬向量编码和K-均值聚类方法,在线性分类器软件包Liblinear上进行分类。为了验证所提出方法的有效性,在三维人体骨架序列行为数据集MSR Action3D上进行了实验。结果表明,与传统的人体行为识别算法相比具有更好的分类精度。

Human behavior recognition has always been the research focus of computer vision, especially with the advent of the Microsoft Kinect sensor, making it easier to get information about the human skeleton. In order to obtain good recognition accuracy on the basis of skeleton model, in this paper, the bag-of-words model and the motion posture descriptor are used based on the three-dimensional skeleton sequence information, and combined with the improved hard vector coding and K-means clustering method, the classification is performed on the Linear classifier software package Lib Linear. In order to verify the effectiveness of the proposed method, experiments were performed on the 3 D human skeleton sequence behavior data set MSR Action3 D. It is found that the classification accuracy is better than the traditional algorithms of human behavior recognition.