【作者】 张娜；

【导师】 田玉敏；

【作者基本信息】 西安电子科技大学 ， 计算机系统结构， 2008， 硕士

【摘要】 步态识别作为一种新兴的生物特征识别技术,目的是通过人走路的姿势实现对个人身份的识别和认证,它是远距离情况下最有潜力的生物特征识别技术之一,因此已经被广泛地应用在智能监控、人机接口和人体行为分析等领域。本文主要研究了两种基于Radon变换的统计矩步态识别算法。在实现算法时,首先采用自适应背景提取算法得到步态序列的背景,再对图像进行预处理以获取二值人体区域,并根据轮廓宽度从每一个步态周期中提取关键帧。第一种算法采用Radon不变矩提取关键帧投影变换的平移、比例和旋转不变量作为步态识别的特征。第二种算法使用本文新提出的Radon速度矩提取步态图像帧之间的时空信息相关性特征作为步态识别的依据。最后使用支持向量机(SVM)分别对这两种特征向量进行训练和识别。通过在两个不同的步态数据库上对两种算法的可行性进行验证,取得了较高的识别率,说明本文采用的方法和提出的新算法都是行之有效的,具有较好的应用前景。更多还原

【Abstract】 As a rising biometric recognition technology, the purpose of gait recognition is to authenticate the identity of people by their posture of walking. It is one of the most potential biometric features. So it has been applied in intelligent supervision, interface between human and computers, and analyzing the behavior of human.Two statistical moments gait recognition methods based on the Radon transform are developed in this thesis. The proposed algorithms are to firstly restore the background of the gait image sequences by adaptive background detecting algorithm. Then images are analyzed to attain binary region of the person. Based on the region width, key frames are separated from every gait cycles. The first method using Radon invariant moments extract the features of key frames by expressing the invariance properties of the images including translation, scale and rotation change. And the second method using Radon velocity moments proposed as a new statistical moment features express the spatial and temporal correlation of projection images. Finally, the two kinds of feature vectors are trained and detected by SVM.The feasibilities of the two algorithms are supported by applying the algorithms on two different gait databases. A greater recognition ratio is achieved, which proves that they can be widely applied in the gait recognition domain in the future.更多还原