基于SSD模型的人脸检测与头部姿态估计融合算法
|
摘要
针对头部姿态估计常用的人脸检测、姿态估计两步串联框架中流程复杂、耦合性高、整体鲁棒性低的问题,提出了一种基于改进SSD模型的人脸检测与头部姿态估计融合算法.通过拓展SSD模型,设计了人脸检测与姿态估计融合网络模型,在多层次卷积特征图上检测人脸,并估计头部姿态;采用端到端训练模式进行模型训练,简化了头部姿态估计任务的处理流程.在Pointing′04和300W-LP数据集上进行了试验.结果表明,本模型能够在满足实时性要求的前提下有效地完成检测任务与估计任务,在两个数据集中的pitch预测平均绝对误差分别达到了4.80°和6.48°,这充分证明了所提出算法的实用性和鲁棒性.
To solve the problems of complex process, high coupling and low robustness in the two-step cascade framework of face detection and pose estimation in common head pose estimation, a fusion algorithm of face detection and head pose estimation was proposed based on the improved SSD model. By expanding the SSD model, a fusion network model of face detection and pose estimation was designed. Face was detected on multi-level convolution feature map, and head pose was estimated. End-to-end training mode was used to train the model, which simplified the processing flow of head pose estimation task. The experiments were completed on Pointing′04 and 300 W-LP datasets. The results show that the proposed model can effectively perform detection and estimation tasks on the premise of satisfying real-time requirements. The pitch prediction average absolute errors in the two datasets are respective 4.80 and 6.48 degrees, which fully proves the practicability and robustness of the proposed algorithm.
To solve the problems of complex process, high coupling and low robustness in the two-step cascade framework of face detection and pose estimation in common head pose estimation, a fusion algorithm of face detection and head pose estimation was proposed based on the improved SSD model. By expanding the SSD model, a fusion network model of face detection and pose estimation was designed. Face was detected on multi-level convolution feature map, and head pose was estimated. End-to-end training mode was used to train the model, which simplified the processing flow of head pose estimation task. The experiments were completed on Pointing′04 and 300 W-LP datasets. The results show that the proposed model can effectively perform detection and estimation tasks on the premise of satisfying real-time requirements. The pitch prediction average absolute errors in the two datasets are respective 4.80 and 6.48 degrees, which fully proves the practicability and robustness of the proposed algorithm.
引文
[ 1 ] JIANG H Z,LEARNED-MILLER E.Face detection with the faster R-CNN[C]//Proceedings of the 12th IEEE International Conference on Automatic Face and Gesture Recognition.New York:IEEE,2017:650-657.
[ 2 ] QIN H W,YAN J J,LI X,et al.Joint training of cascaded CNN for face detection[C]//Proceedings of the 29th IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE Computer Society,2016:3456-3465.
[ 3 ] LEE S,SAITOH T.Head pose estimation using convolutional neural network[C]//Proceedings of the 7th International Conference on IT Convergence and Security.Heidelberg:Springer Verlag,2017:164-171.
[ 4 ] LIU W,ANGUELOV D,ERHAN D,et al.SSD:single shot multibox detector[C]//Proceedings of the Euro-pean Conference on Computer Vision.Heidelberg:Springer,2016:21-37.
[ 5 ] STIEFELHAGEN R.Estimating head pose with neural networks-results on the pointing04 ICPR workshop eva-luation data[C]//Proceedings of the ICPR Workshop Visual Observation of Deictic Gestures,2004.
[ 6 ] OSADCHY M,CUN Y L,MILLER M L.Synergistic face detection and pose estimation with energy-based models[J].Journal of Machine Learning Research,2007,8(1):1197-1215.
[ 7 ] REN S Q,HE K M,GIRSHICK R,et al.Faster R-CNN:towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Ana-lysis & Machine Intelligence,2017,39(6):1137-1149.
[ 8 ] PARKHI O M,VEDALDI A,ZISSERMAN A.Deep face recognition[C]//Proceedings of the British Machine Vision Conference,doi:10.5244/C.29.41.
[ 9 ] GENG X,XIA Y.Head pose estimation based on multivariate label distribution[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Re-cognition.New York:IEEE,doi:10.1109/CVPR.2014.237.
[10] LI Y L,WANG S J,DING X Q.Person-independent head pose estimation based on random forest regression[C]//Proceedings of 2010 17th IEEE International Conference on Image Processing.New York:IEEE,doi:10.1109/ICIP.2010.5652915.
[ 2 ] QIN H W,YAN J J,LI X,et al.Joint training of cascaded CNN for face detection[C]//Proceedings of the 29th IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE Computer Society,2016:3456-3465.
[ 3 ] LEE S,SAITOH T.Head pose estimation using convolutional neural network[C]//Proceedings of the 7th International Conference on IT Convergence and Security.Heidelberg:Springer Verlag,2017:164-171.
[ 4 ] LIU W,ANGUELOV D,ERHAN D,et al.SSD:single shot multibox detector[C]//Proceedings of the Euro-pean Conference on Computer Vision.Heidelberg:Springer,2016:21-37.
[ 5 ] STIEFELHAGEN R.Estimating head pose with neural networks-results on the pointing04 ICPR workshop eva-luation data[C]//Proceedings of the ICPR Workshop Visual Observation of Deictic Gestures,2004.
[ 6 ] OSADCHY M,CUN Y L,MILLER M L.Synergistic face detection and pose estimation with energy-based models[J].Journal of Machine Learning Research,2007,8(1):1197-1215.
[ 7 ] REN S Q,HE K M,GIRSHICK R,et al.Faster R-CNN:towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Ana-lysis & Machine Intelligence,2017,39(6):1137-1149.
[ 8 ] PARKHI O M,VEDALDI A,ZISSERMAN A.Deep face recognition[C]//Proceedings of the British Machine Vision Conference,doi:10.5244/C.29.41.
[ 9 ] GENG X,XIA Y.Head pose estimation based on multivariate label distribution[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Re-cognition.New York:IEEE,doi:10.1109/CVPR.2014.237.
[10] LI Y L,WANG S J,DING X Q.Person-independent head pose estimation based on random forest regression[C]//Proceedings of 2010 17th IEEE International Conference on Image Processing.New York:IEEE,doi:10.1109/ICIP.2010.5652915.