項目中需要做人眼檢測,目前主要有adaboost,,svm,基於相關算法的人眼檢測,膚色分割瞳孔定位等算法。由於人眼特徵數目比較少, adaboost的誤檢率和精準度都不行,而SVM比較慢,膚色空間等瞳孔定位太不穩定,反倒像ASEF,MOSSE等算法測試的效果還比較好,但是在閉眼,眼鏡遮擋上還是不行,但是由於相關算法在頻域中處理速度較快,因此可以考慮多訓練幾個模板。
參考文章:
PTIMIZED CORRELATION FILTERS FOR SIGNAL PROCESSINGDS Bolme, JR Beveridge, BA Draper
US Patent App. 12/797,241
DS Bolme, Ph.D. Dissertation
COLORADO STATE UNIVERSITY
Average of synthetic exact filters
DS Bolme, BA Draper, JR Beveridge
Computer Vision and Pattern Recognition, 2009. CVPR 2009.
Visual object tracking using adaptive correlation filters
DS Bolme, JR Beveridge, BA Draper, YM Lui
Computer Vision and Pattern Recognition (CVPR), 2010
Simple real-time human detection using a single correlation filter
DS Bolme, YM Lui, BA Draper, JR Beveridge
Performance Evaluation of Tracking and Surveillance (PETS-Winter), 2009
http://www.cs.colostate.edu/~vision/ocof_toolset_2012/index.php