红外序列图像中基于形状的人体检测

SHAPE-BASED HUMAN DETECTION IN INFRARED IMAGE SEQUENCES

对红外序列图像中的人体检测问题进行了研究,提出一种新的人体检测方法.首先采用自适应高斯混合模型对序列图像中背景进行建模,在准确分割出前景运动目标的基础上,提出了一种新的人体形状表达模型,充分考虑了多个人体发生粘连或互相遮挡的情况,并用亮度投影的方法对其进行分离;以人体表达模型作为输入向量,构建支持向量机(SVM,Support Vector Machine)对人体进行分类判别.不同红外视频序列的检测结果表明了所提出算法在单个人体和多人体情况下均具有较好的鲁棒性和可行性.

The human detection problem in infrared image sequences was studied, and a novel detecting approach was presented. GMM （Gauss mixture model ） was first adopted to construct a background model. And then on the basis of accurately segmenting the forward objects, a shape-based human representing model was designed. By taking account of occlusions and merging among multi-body, the intensity projection curve was applied to separate single ones. By using human shape models as input vectors, a SVM （support vector machine） was constructed to classify and identify the human bodies. Experimental results on different infrared video sequences show that the proposed method is robust and feasible in single body and multi-body cases.