局部证据RBF人体行为高层特征自相似融合识别研究

Local Evidence RBF Algorithm Based High-level Characteristic Self-similarity Fusion for Human Behavior Recognition

针对传统人体动作识别算法,往往重点解决某一类行为识别,不具有通用性的问题,提出一种局部证据RBF人体行为高层特征自相似融合识别算法。首先,借用随时间变化的广义自相似性概念,利用时空兴趣点光流场局部特征提取方法,构建基于自相似矩阵的人体行为局部特征描述;其次,在使用SVM算法进行独立个体行为识别后,利用所提出的证据理论RBF(Radial Basis Function)高层特征融合,实现分类结构优化,从而提高分类准确度;仿真实验表明,所提方案能够明显提高人体行为识别算法效率和识别准确率。

The traditional human action recognition algorithm tends to focus on solving a certain behavior recognition, it cannot be generalized. So, this paper put forward a kind of Local evidence RBF algorithm based high-level characteristic self similarity fusion for human behavior recognition. Firstly, the time-dependent generalized self similarity concept and the spario--temporal interest point optical flow based local features extraction method were used to construct the human behavior description based on self similar matrix. Secondly, after independent individual behavior recognition in the use of SVM algorithm, the evidence theory based high level feature fusion was used to realize the optimization for classification of structure, which can improve the accuracy of classification. Simulation results show that the proposed scheme can significantly improve the efficiency and accuracy for human action recognition.