仿生模式识别理论在45°视角的步态识别中的应用  

APPLICATION OF BIOMIMETIC PATTERN RECOGNITION IN GAIT RECOGNITION WITH 45° VIEW ANGLE

作　　者：李一波[1] 孟迪[1] 

机构地区：[1]沈阳航空航天大学自动化学院,辽宁沈阳110136

出　　处：《计算机应用与软件》2015年第1期190-193,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(61103123)

摘　　要：结合仿生模式识别理论,提出一种利用多权值神经元网络来进行步态识别的方法。提取同一时刻左右小腿关节点运动的速度场和关节角度,构成特征向量,利用多权值神经元网络形成的复杂几何体在特征空间中构造不同人步态特征的最小覆盖,从而达到步态识别的目的。相关实验表明,该方法在保证较高的识别率的同时,可以有效提高拒识率。与传统方法相比,误识率明显下降。In combination with biomimetic pattern recognition theory, we present a method which uses multi-weight neural networks to recognise the gait. It extracts at the same time the velocity field and joint angle of two cruses to form the eigen vector, and utilises the complicated geometry generated with multi-weight neural networks to construct the minimum covering of different peoples＇ gait features in feature space. Relevant experiments show that the rejection rate can be effectively improved by the method of BPR while the recognition rate keeps high. Compared with traditional method, the misrecognition rate is reduced remarkably.

关 键 词：仿生模式识别 超香肠神经元 多权值神经网络 步态识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]