基于特征融合和神经网络对步态识别仿真研究

Gait Recognition Algorithm Based on Feature Fusion and Neural Network

研究人体步态识别问题,根据人体步态变化特点,提出一种基于特征融合和神经网络的步态识别算法。首先采用时域差分法对运动人体轮廓进行分割,然后分别提取空间特征和频率特征,将两步态特征融合在一起,最后将得到的融合特征向量输入到神经网络进行学习,从而实现步态的分类和识别。在CASIA步态数据库上进行对比仿真,仿真结果表明,方法不仅克服了单一特征提取方法存在的缺陷,同时提高了步态识别正确率。

According to the characteristics of human gait changes, this paper proposed a gait recognition algorithm based on feature fusion and neural network. Firstly, time domain finite difference method was used to obtain extract body silhouette and extract spatial features and frequency characteristics. Then two kinds of features are fused. Final- ly, the fusion feature vectors were input into the neural network for learning, so as to realize the classification and recognition of gaits. The algorithm was tested by CASIA gait database. The simulation results show that the proposed method not only overcomes the defects of single feature extraction method, but also improves gait recognition correct rate.