摘要
根据综合利用步态的静态和动态信息的思想,结合不变矩描述图像几何特性的功能,从步态序列提取不变矩作为步态特征进行识别。采用傅立叶级数描述步态图像序列人体轮廓不变矩的变化,利用遗传算法搜索傅立叶级数的系数,最后再用k近邻分类器对不变矩变化的幅度信息分类。在CMU步态数据库上进行的实验,达到了90%以上的识别率。结果表明,该方法具备很高识别性能,能较好地利用步态的静态和动态信息。
According to the idea of combining static components and dynamic components from the walking way, as the moment invariants may represent geometrical traits in images, they were extracted as gait features from a subsequent gait series. The moment invariants of human silhouettes were represented by the Fourier series. A genetic algorithm was deployed to obtain the Fourier coefficients. The magnitudes of the coefficients were classified through the kNN classifier. The recognition restdts with the proposed scheme in the CMU gait database have recognition rates of more than 90% and show that it has achieved a high performance and effectively made use of two kinds of contents.
出处
《计算机应用》
CSCD
北大核心
2007年第4期922-924,928,共4页
journal of Computer Applications
基金
重庆市自然科学基金资助项目(CSTC2006BB2155)
关键词
特征提取
步态识别
不变矩
遗传算法
KNN分类器
feature extraction
gait-based recognition
moment invariants
genetic algorithm（GA）
kNN classifier
