【摘要】 为了生成静态图像中人体的轮廓线,提出了一种基于骨架匹配的轮廓线提取方法。在图像中找到可以表征人体骨架的骨架关键点,通过骨架关键点,利用图像中各像素点所包含的边缘信息,结合人体形态学先验知识,构建了隐马尔科夫模型来匹配人体上对应骨架关键点的边缘关键点,少量的边缘关键点即可生成人体大致轮廓。利用Hermite插值方法连接边缘关键点获取人体轮廓线。通过相关实验结果表明,提出的方法在较为简单的场景中可以提取出静态图像中人体轮廓,获得的轮廓也符合人体形状先验。与已有的方法相比,在速度上有了很大的提高,并在一些较为复杂的场景中取得了更好的效果。更多还原

【Abstract】 To extract human contour in still images, we propose a human contour extraction method based on skeleton matching. Skeleton key points are points representing human skeleton in images. Optimal human contour key points corresponding to skeleton key pointscanbe extracted by constructing hidden Markov modelthrough skeleton key points and image pixels’ edge information combined with human body morphology as prior knowledge. Few human contour key points can generate reasonable human contour. The Hermiteinterpolation is usedto connect these key points to obtain human contour. Experimentresults show thatour method can extract reasonable human contour in simple scene, andhas higher speed and achieves better results in complicated scenecompared to previous methods.更多还原