计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (7): 154-159.DOI: 10.3778/j.issn.1002-8331.1509-0268

• 模式识别与人工智能 • 上一篇    下一篇

结合运动特征的AdaBoost层次增强算法

罗富丽,李佳田   

  1. 昆明理工大学 国土资源工程学院,昆明 650093
  • 出版日期:2017-04-01 发布日期:2017-04-01

AdaBoost hierarchical enhancement algorithm combined with moving features

LUO Fuli, LI Jiatian   

  1. Faculty of Land Resources Engineering, Kunming University of Science and Technology, Kunming 650093, China
  • Online:2017-04-01 Published:2017-04-01

摘要: 随着高分辨率传感器成为视频获取的主流,AdaBoost算法所面临的主要问题是置入级联分类器待检测窗口数过多。提出一种AdaBoost人脸检测层次增强算法,以加快人脸检测速度。从整体目标运动与局部人脸运动两个层次出发,以矩形块为计算单位,根据相关性原则,提取运动目标区域;以运动特征为基础,结合主成分分析获得运动特征子空间;通过子空间投影得到候选人脸窗口集合。对比实验表明,在640×480以及1?280×720视频帧中,该算法具有较高的子窗口置入率和稳定的检测精度,平均检测速度分别为28?f/s和6?f/s,适用于实时人脸检测。

关键词: 人脸检测, AdaBoost算法, 运动特征, 光流, 主成分分析

Abstract: With the main trend of high-revolution remoter in video capture, the fundamental issue of AdaBoost algorithm is the amount of detection windows placed in the cascade classifiers. The hierarchical duplex AdaBoost algorithm is proposed to accelerate the detection speed. Based on the global target motion and local face feature, it extracts the target motion ranges from the relativity based on the patch, gets the sub-space of feature combining with Principal Component Analysis(PCA) based on the motion features, obtains the set of candidate windows by the sub-space projection. Compared with experiment results, on the video frame of 640×480 and 1?280×720, the hierarchical duplex AdaBoost algorithm has higher sub-window placement rate and stable detection speed, and the average detection speed is 28 f/s and 6 f/s respectively. So it is adapted to the real-time detection.

Key words: face detection, AdaBoost algorithm, moving feature, optical flow, Principal Component Analysis(PCA)