摘要： 针对夜间行人的身份识别问题，结合步态轮廓形状特征及模型投影特征，提出一种红外图像中的混合步态识别方法。采用Radon变换获取步态图像的形状特征，建立3D自适应人体模型，给出基于3D模型的跟踪方法，以获得步态模型特征，并利用SVM进行分类。实验结果表明，该方法具有一定的鲁棒性，识别率可达95.28%。

关键词: 步态识别, 支持向量机, 行人跟踪, 边缘匹配误差, 区域匹配误差

Abstract: Aiming at the identity recognition problem for pedestrian in night, combined with gait outline character and model projection character, a hybrid feature combining the shape information obtained by Radon transform and the walking style information obtained by human tracking is presented. The 3D model tracking method is given. Experimental results show the method is robust for occlusion and human condition changing. SVM is used for classification. The recognition rate can be 95.28%.

Key words: gait recognition, Support Vector Machine(SVM), human tracking, boundary matching error, region matching error

中图分类号: 

• N945