CAJ | 학술논문

针对从多帧步态中更有效提取步态特征的问题,提出了一种基于嵌入式隐马尔可夫模型的步态识别算法.首先采用背景减除方法提取出入体的侧影轮廓,通过分析轮廓宽度向量的自相关性计算出步态的周期,并得到平均步态能量图.接着利用二维离散余弦变换获得平均步态能量图的空间特征信息,然后把能量图的观测块转化为观测向量实现了步态识别.最后运用最近邻法在两个不同的数据库上进行算法验证,实验结果表明该算法具有较好的识别性能.
침대종다정보태중경유효제취보태특정적문제,제출료일충기우감입식은마이가부모형적보태식별산법.수선채용배경감제방법제취출입체적측영륜곽,통과분석륜곽관도향량적자상관성계산출보태적주기,병득도평균보태능량도.접착이용이유리산여현변환획득평균보태능량도적공간특정신식,연후파능량도적관측괴전화위관측향량실현료보태식별.최후운용최근린법재량개불동적수거고상진행산법험증,실험결과표명해산법구유교호적식별성능.
A gait recognition algorithm based on embedded hidden Markov model (e-HMM) is proposed for the problem of extracting features efficiently from a sequence of gait frames. First, body silhouette extraction is achieved by background subtraction method. Gait period is calculated through analyzing the silhouette-width-vector autocorrelation and the mean gait energy image (GEl) is obtained. Then, spatial feature information of the mean GEl is obtained by two dimensional discrete cosine transform (2D-DCT). The mean GEI observation block is transformed into observation vector and gait recognition is performed. In the end, the nearest-neighbor method is adopted to confirm the proposed algorithm using two different gait databases. Experimental results show that the algorithm has higher recognition performance.