计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (21): 190-194.DOI: 10.3778/j.issn.1002-8331.1606-0319

• 图形图像处理 • 上一篇    下一篇

融合HOG和颜色特征的人体姿态估计新算法

沈建冬,陈  恒   

  1. 西京学院 控制工程学院,西安 710123
  • 出版日期:2017-11-01 发布日期:2017-11-15

New human pose estimation algorithm based on HOG and color features

SHEN Jiandong, CHEN Heng   

  1. College of Control Engineering, Xijing University, Xi’an 710123, China
  • Online:2017-11-01 Published:2017-11-15

摘要: 为解决现有人体姿态估计算法在处理光照条件很差或颜色对比度很低的待处理图像时估计准确度较低的问题,利用梯度方向直方图(HOG)和颜色特征建立了一种的基于可能性C(PCM)聚类算法部位外观模型,提出了一种新的融合HOG特征和颜色特征的人体姿态估计算法。算法根据待处理图像自动选择部位外观模型,若图像的光照条件和颜色对比度都较好则选择现有的基于HOG和颜色特征融合的部位外观模型,否则选择基于PCM聚类算法的部位外观模型。仿真实验表明所建立的部位外观模型能更准确地描述光照条件很差或颜色对比度很低的图像中下真实人体部位的外观,提出的人体姿态估计算法对各种类型的待处理图像均能得到准确度更高的估计结果。

关键词: 人体姿态估计, 部位外观模型, 梯度方向直方图, 颜色, 可能性C聚类算法

Abstract: The existing human pose estimation algorithm always get the lower accuracy for those images with very poor light conditions and low color contrast. For solving the problem, a more suitable part appearance model based on the Possibilistic C-Means(PCM) clustering algorithm is built by using Histogram of Oriented Gradient(HOG) and color features, and a new human pose estimation algorithm based on the fusion of HOG and color features is proposed. The part appearance model is selected automatically according to the image to be processed, the existing part appearance model based on the fusion of HOG and color features is selected if both the light conditions and color contrast are good, and otherwise the part appearance model based on PCM clustering algorithm is chosen. Simulation results show that the established part appearance model can represent the appearance of real human part, which is from images with very poor light conditions and low color contrast, more accurately, the proposed human posture estimation algorithm can get more accurate estimation results for various types of images to be processed.

Key words: human pose estimation, part appearance model, histogram of oriented gradient, color, possibilistic C-means clustering algorithm