自动化学报Issue(8)：1623-1634,12.DOI:10.3724/SP.J.1004.2014.01623
自适应融合颜色和深度信息的人体轮廓跟踪
Adaptively Combining Color and Depth for Human Body Contour Tracking
摘要
Abstract
In this paper, we present a novel human body contour tracking method, which combines color and depth cues of RGB-D images adaptively in the level set framework. We model the body object by the active contour. A superpixel-based locally adaptive weight map is designed to determine the importance of the depth cue in the evolution of the active contour. The depth-based external forces for the active contour are derived from the gradient vector flow (GVF) generated from the edges of the depth image and the confidence map generated from the depth models of the ob ject/background, while the color-based external force is derived from the confidence map generated from the color models of the object/background. The three external forces are integrated by the adaptive weight to drive the active contour to evolve to the boundary of the object. To obtain a more accurate contour and to avoid error drifting, we propose two simple but effective algorithms based on the two properties of the human body surface in the depth image to refine the tracking result of the level set method. Experimental results demonstrate that our tracking method behaves more robustly and accurately than the latest depth-based body contour extraction method and the color-based contour tracking method on the datasets acquired in indoor environments.关键词
轮廓跟踪/人体跟踪/活动轮廓/水平集Key words
Contour tracking/human tracking/active contour/level set引用本文复制引用
徐玉华,田尊华,张跃强,朱宪伟,张小虎..自适应融合颜色和深度信息的人体轮廓跟踪[J].自动化学报,2014,(8):1623-1634,12.基金项目
国家重点基础研究发展计划(973计划)(2013CB733100)资助Supported by National Basic Research Program of China (973 Program)(2013CB733100) （973计划）
