一种基于改进ACF特征的手部检测方法
|
摘要
如何在存在噪声和类肤色背景的环境中进行高效精确的手部检测,是手部检测研究的一大问题.提出一种基于改进ACF的检测方法.该算法在基于多色彩空间肤色模型和边缘直方图基础上对ACF特征进行改进,以多色彩空间肤色模型来突出肤色物体与非肤色物体之间的差异,以边缘直方图来描述物体间的边界信息.同时为提高算法性能,对物体检测算法框架进行了改进.一方面改进特征计算过程,对于每个图像只进行一次特征计算;另一方面利用Edge Boxes获取候选窗口,以此减少候选窗口的数量.最后使用Xgboost对每个候选窗口对应的特征进行判别.实验证明,在存在高斯噪声和受人脸干扰的情况下,该方法可以有效地进行手部检测.
How to detect hand in the presence of noise and skin-like background is a big problem in hand detection. Aiming at this problem,a detection method based on improved ACF feature is proposed in this paper. Firstly,the ACF feature is improved based on the HOE and the multi-colorspace skin model,the multi-colorspace skin model is used to highlight the difference between the skin color object and the non-skin color object,and the HOE is used to describe the boundary information between objects. In order to improve the performance of the algorithm,this paper improves the frame of object detection algorithm. On the one hand,the feature calculation is adjusted,and for each image,the algorithm performs only one feature calculation. On the other hand,the Edge Boxes algorithm is used to get the proposals,so as to reduce the number of candidate windows. And then,the corresponding features of each proposal would be judged by Xgboost. Finally,it is proved by experiments in the presence of noise and skin-like background,the method can effectively carry out hand detection.
How to detect hand in the presence of noise and skin-like background is a big problem in hand detection. Aiming at this problem,a detection method based on improved ACF feature is proposed in this paper. Firstly,the ACF feature is improved based on the HOE and the multi-colorspace skin model,the multi-colorspace skin model is used to highlight the difference between the skin color object and the non-skin color object,and the HOE is used to describe the boundary information between objects. In order to improve the performance of the algorithm,this paper improves the frame of object detection algorithm. On the one hand,the feature calculation is adjusted,and for each image,the algorithm performs only one feature calculation. On the other hand,the Edge Boxes algorithm is used to get the proposals,so as to reduce the number of candidate windows. And then,the corresponding features of each proposal would be judged by Xgboost. Finally,it is proved by experiments in the presence of noise and skin-like background,the method can effectively carry out hand detection.
引文
[1]Bhuyan M K,Neog D R,Kar M K.Hand pose recognition using geometric features[C].Institute of Electrical and Electronics Engineers International Conference on Signal and Image Processing Applications,Kuala Lumpur,2011:378-383.
[2]Stergiopoulou E,Sgouropoulos K,Nikolaou N,et al.Real time hand detection in a complex background[J].Engineering Applications of Artificial Intelligence,2014,35(2):54-70.
[3]Sangineto E,Cupelli M.Real-time viewpoint-invariant hand localization w ith cluttered backgrounds[J].Image&Vision Computing,2012,30(1):26-37.
[4]Wang Kai,Wu Min,Yao Hui,et al.Moving target detection method based on multi-frame background subtraction and double-threshold[J].Journal of Chinese Computer Systems,2017,38(1):179-183.
[5]Mei K,Xu L,Li B,et al.A real-time hand detection system based on multi-feature[J].Neurocomputing,2015,158(C):184-193.
[6]Viola P,Jones M J.Robust real-time object detection[C].International Workshop on Statistical and Computational Theories of Vision-M odeling,Learning,Computing,and Sampling,Vancouver,Canada,2001,57(2):87-110.
[7]Forsyth D.Object detection with discriminatively trained part-based models[J].Institute of Electrical and Electronics Engineers Transactions on Pattern Analysis&M achine Intelligence,2010,32(9):1627-1645.
[8]Yang B,Yan J,Lei Z,et al.Aggregate channel features for multiview face detection[C].Institute of Electrical and Electronics Engineers International Joint Conference on Biometrics,Clearw ater,Florida.2014:1-8.
[9]Girshick R,Donahue J,Darrell T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C].Institute of Electrical and Electronics Engineers Conference on Computer Vision and Patern Recognition,Columbus,Ohio,2014:580-587.
[10]Cheng M M,Zhang Z,Lin W Y,et al.Bing:binarized normed gradients for objectness estimation at 300fps[C].Institute of Electrical and Electronics Engineers Conference on Computer Vision and Pattern Recognition,Columbus,Ohio,2014:3286-3293.
[11]Zitnick C L,Dollár P.Edge Boxes:locating object proposals from edges[M].Computer Vision-European Conference on Computer Vision 2014,Springer International Publishing,2014:391-405.
[4]王凯,吴敏,姚辉,等.多帧背景差与双门限结合的运动目标检测方法[J].小型微型计算机系统,2017,38(1):179-183.
[2]Stergiopoulou E,Sgouropoulos K,Nikolaou N,et al.Real time hand detection in a complex background[J].Engineering Applications of Artificial Intelligence,2014,35(2):54-70.
[3]Sangineto E,Cupelli M.Real-time viewpoint-invariant hand localization w ith cluttered backgrounds[J].Image&Vision Computing,2012,30(1):26-37.
[4]Wang Kai,Wu Min,Yao Hui,et al.Moving target detection method based on multi-frame background subtraction and double-threshold[J].Journal of Chinese Computer Systems,2017,38(1):179-183.
[5]Mei K,Xu L,Li B,et al.A real-time hand detection system based on multi-feature[J].Neurocomputing,2015,158(C):184-193.
[6]Viola P,Jones M J.Robust real-time object detection[C].International Workshop on Statistical and Computational Theories of Vision-M odeling,Learning,Computing,and Sampling,Vancouver,Canada,2001,57(2):87-110.
[7]Forsyth D.Object detection with discriminatively trained part-based models[J].Institute of Electrical and Electronics Engineers Transactions on Pattern Analysis&M achine Intelligence,2010,32(9):1627-1645.
[8]Yang B,Yan J,Lei Z,et al.Aggregate channel features for multiview face detection[C].Institute of Electrical and Electronics Engineers International Joint Conference on Biometrics,Clearw ater,Florida.2014:1-8.
[9]Girshick R,Donahue J,Darrell T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C].Institute of Electrical and Electronics Engineers Conference on Computer Vision and Patern Recognition,Columbus,Ohio,2014:580-587.
[10]Cheng M M,Zhang Z,Lin W Y,et al.Bing:binarized normed gradients for objectness estimation at 300fps[C].Institute of Electrical and Electronics Engineers Conference on Computer Vision and Pattern Recognition,Columbus,Ohio,2014:3286-3293.
[11]Zitnick C L,Dollár P.Edge Boxes:locating object proposals from edges[M].Computer Vision-European Conference on Computer Vision 2014,Springer International Publishing,2014:391-405.
[4]王凯,吴敏,姚辉,等.多帧背景差与双门限结合的运动目标检测方法[J].小型微型计算机系统,2017,38(1):179-183.