摘要
根据人机交互中手势控制系统的要求,提出一种基于平均邻域最大化(ANMM)算法的静态手势识别方法。将获得的二值化图像轮廓归一化到固定的解析度,构成多维向量,使用ANMM算法对同质相邻与异质相邻向量进行训练,计算出投影方向矩阵。将样本降维处理后,计算其在降维空间内与同质相邻和异质相邻向量的距离,判别样本所属分类。实验结果证明,该方法对静态手势的识别率可达90%以上。
According to the requirement of hand posture control system in the field of human machine interaction, Average Neighborhood Margin Maximization（ANMM） algorithm is applied to static hand posture and human body posture recognition. It normalizes the binary image contour into a fixed resolution, constitutes a multi-dimensional vector, uses the ANMM algorithm training the homogeneity and heterogeneity neighboring vectors, and then projection direction matrix is calculated. After reducing the dimension of the sample, calculating the distance of the adjacent homogeneity and heterogeneity vectors, the samples are classified. Results show that this algorithm has a recognition rate of 90% for the static hand posture recognition.
出处
《计算机工程》
CAS
CSCD
2012年第9期19-20,35,共3页
Computer Engineering
基金
国家自然科学基金资助项目(61103113)
关键词
手势识别
平均邻域最大化
特征提取
相邻同质
相邻异质
降维
hand posture recognition
Average Neighborhood Margin Maximization（ANMM）
feature extraction
adjacent homogeneity
adjacentheterogeneity
dimension reduction
