摘要
针对图像光照的变化对静态头部姿态估计的影响,该文提出一种基于有向梯度直方图和主成分分析的姿态特征,并利用SVM分类器进行分类。该算法分别在CMU姿态、光照、表情数据库和CVL人脸图像库上进行了测试。实验表明,即使图像光照变化很大,该算法仍可准确地估计头部姿态,识别率达到90%以上。
To solve head pose estimation problem under different illumination and expression, a new pose feature is proposed, which is based on Histogram of Oriented Gradient(HOG) feature and Principle Component Analysis(PCA), and SVM classifier is used to classify different poses. The algorithm is tested based on CMU pose, illumination and expression database and CVL database. Experiment shows that the algorithm can estimate poses accurately under different illumination, and correct ratio arrives at 90%.
出处
《计算机工程》
CAS
CSCD
北大核心
2008年第10期16-18,共3页
Computer Engineering
关键词
姿态估计
有向梯度直方图
光照
pose estimation
Histogram of Oriented Gradient(HOG)
illumination