基于部位分割的人体再识别方法

Human Re-identification Based on Part Segmentation

人体再识别是非重叠视域多摄像机视频分析过程中需解决的难点问题。提出一种基于部位分割的人体再识别方法,首先基于深度骨骼点实现人体部位分割,对同一人体多帧图像分割后的所有部位使用评分的策略选择最优帧,将全局颜色特征和HOG特征分配不同的权重,融合特征建立人体目标模型,使用EMD(Earth Move’s Distance)距离度量目标间的相似性。在Kinect REID和BIWI RGBD-ID数据库上实验表明,该算法具有较高的鲁棒性和识别率。

Human re-identification is a difficult problem to solve in process of video analysis of non-overlapping multi-camera surveillance system. A new algorithm of human re-identification is proposed on the basis of human part segmentation. Based on the depth of bone points to achieve the human body segmentation, the optimal key frame is selected by using the scoring strategy for all parts of the same human multi-frame image segmentation;the different weights for the global color feature and the HOG feature are assigned;all the characteristics to establish a human target model are combined;and the EMD(Earth Mover’s Distance) distance is used to determine the similarity between the targets. The effectiveness is validated on Kinect REID and BIWI RGBD-ID datasets which show that the proposed method has strong robustness and higher recognition rate.