基于视角相关角度分量特征的步态识别-计算机应用技术专业论文.docxVIP

II II 离的平均值，并把该平均值分解为水平和垂直两个分量作为当前帧的 角度分量特征，再计算第一帧到当前帧的角度分量特征的平均值作为 最终的步态特征。由于该方法考虑到水平和垂直两个方面对识别的影 响，因此相对角度特征具有更好的分类能力，尤其提高了在非侧面视 角下的识别率。 最后，在步态识别过程中，本文采用基于视角的多分类器融合的 识别方法，根据视角检测过程得到的视角选择在该视角下最优的分类 器进行步态识别。本文融合使用线性时间归一化（LTN）匹配法和常 用的 k 近邻法（KNN）进行步态识别，两种方法优势互补，既发挥了 KNN 方法在各个视角识别的相对稳定性，又发挥了 LTN 方法在侧面视 角下识别率比较高的优势，最终实现在每个视角都能够选择最优的识 别方法从而达到最好的识别效果。 关键字：生物特征识别；步态识别；角度分量特征；多分类器融合； 视角检测。 III III Abstract The gait recognition is one new research topic of biometrics recognition in the field of computer vision and intelligent video surveillance, which makes use of human walking pattern to recognize and identify. Compared with other biometrics recognition such as fingerprints recognition, iris recognition and face recognition, gait recognition has advantage of non-contact, non-infringe, easy to collect, as well as remote sensing. Based the features above, gait recognition has vast application prospects and huge economic value in access control system, man-machine interaction and medical diagnoses fields. A whole gait recognition system have three main parts: moving object detection, gait feature extraction and classification and identification. The article learns and studies the domestic and foreign researches on gait, and discusses some technologies about the three key parts of gait. First of all, the article summarizes some gait recognition technologies, and discusses the contribution and advantages of each method. It also simply discusses the popular gait recognition method, current research status and development tendency. Secondly, at the moving object detection part, the article adopts PAGE PAGE IV Gaussian Mixed Model based on chromaticity coordinates to boost detection rate on low-contrast pixel and ensure the accuracy of gait detection, which is very important for angle of view detection. Then, according to the image formation principle in Graphics, calculate the visual angle in three-dimensional space. The visual angle helps selecting a better classification