Abstract: A novel method is proposed for recognition of human action based on improved boosting RBF neural network.In the proposed method，normalized motion history image for motion representation is valued.Statistical descriptions are then computed from motion history image using Zernike moment-based features，then adaboost method is proposed for adaptive select feature.Then RBF neural network is used to class the human action.In order to improve the precision of the RBF neural network for recognition of human action，a weight-adjusting-based method is proposed to improve Boosting method.Experiment results have shown good recognition performance of our method.

Key words: Zernike moments, recognition of human action, Boosting method, motion history image

摘要： 提出一种基于Boosting RBF神经网络的人体行为识别方法，该方法利用规范化的运动历史图像（MHI）进行图像序列表示，从中提取Zernike矩的统计描述特征，然后提出Adaboost算法自适应地选择图像序列的特征作为RBF神经网络的输入，为了进一步提高神经网络的泛化能力，采用一种调整权值分布，限制权重扩张的改进的Boosting方法，分类器以加权投票方式进行分类决策。实验结果表明，提出的方法能够有效地识别人体运动类别。

关键词: Zernike矩, 人体行为识别, boosting 算法, 运动历史图像