基于聚类RBF神经网络的人体行为识别

摘要

提出基于聚类RBF神经网络的人体行为识别方法.通过基于单模态高斯背景模型的背景差分法提取动作轮廓；采用基于中心距的傅里叶描述子,对图像轮廓线进行处理,降低了特征的维数；利用谱聚类算法提取行为序列的关键特征向量,采用改进的基于聚类的RBF神经网络进行行为识别.仿真实验表明,该方法能有效识别人体行为类别,应用效果满足实际要求.%A human behaviour recognition method based on clustering RBF neural network is proposed in this paper. The motion contour is extracted by the background subtraction which is based on single mode Gauss background model. A kind of Fourier descriptor based on centre distance is adopted for processing the contour lines of image, and the feature dimensions are reduced. Spectral clustering method is used to extract the key characteristic vectors of behaviour sequence, and an improved clustering-based RBF neural network is adopted to recognise human behaviour. Simulation experiments show that this method can effectively recognise the category of human behaviours, and the application results meet the practical requirements as well.