基于TINY-YOLO的嵌入式人脸检测系统设计
|
摘要
基于深度学习的人脸检测算法对于人脸检测性能有了很大的提升,但是大多数算法无法部署在移动端上。如MTCNN等算法虽然可以部署在嵌入式系统,但是在实际的应用场景中,有时不能满足需求。将TINY-YOLO用于人脸检测,结合NCNN框架,实现了在移动端的部署及商用。精度上相比于MTCNN有较大的提升,在实际的应用场景中表现优异,并且算法运行时间不会随着人脸数量的增加而增加,更适合于多人脸的嵌入式环境下人脸检测。
Depth learning based face detection algorithm has greatly improved the performance of face detection,but most of the algorithms can not be deployed on the mobile terminals.Although algorithms such as MTCNN can be deployed in embedded systems,they sometimes can not meet the requirements in actual application scenarios.In this paper,TINY-YOLO is used for face detection,and the NCNN framework is used to implement the deployment and business in mobile terminals.Compared with MTCNN,the precision of the algorithm is improved greatly,and it performs well in practical application scenarios.Moreover,the running time of the algorithm does not increase with the increase of the number of faces,so it is more suitable for face detection in multi-face embedded environment.
Depth learning based face detection algorithm has greatly improved the performance of face detection,but most of the algorithms can not be deployed on the mobile terminals.Although algorithms such as MTCNN can be deployed in embedded systems,they sometimes can not meet the requirements in actual application scenarios.In this paper,TINY-YOLO is used for face detection,and the NCNN framework is used to implement the deployment and business in mobile terminals.Compared with MTCNN,the precision of the algorithm is improved greatly,and it performs well in practical application scenarios.Moreover,the running time of the algorithm does not increase with the increase of the number of faces,so it is more suitable for face detection in multi-face embedded environment.
引文
[1]Girshick R,Donahue J,Darrell T,et al.Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation[C]//2014 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).IEEE Computer Society,2014
[2]Girshick R.Fast R-CNN[J].Computer Science,2015
[3]Ren S,He K,Girshick R,et al.Faster R-CNN:towards real-time object detection with region proposal networks[C]//International Conference on Neural Information Processing Systems,2015
[4]Liu W,Anguelov D,Erhan D,et al.SSD:Single Shot MultiBox Detector[C]//European Conference on Computer Vision.Springer International Publishing,2016:21-37
[5]Redmon J,Divvala S,Girshick R,et al.You Only Look Once:Unified,Real-Time Object Detection[C]//Computer Vision&Pattern Recognition,2016
[6]Redmon J,Farhadi A.YOLO9000:Better,Faster,Stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR),2016:6517-6525
[7]Redmon J,Farhadi A.YOLOv3:An Incremental Improvement[J].2018
[8]Wang H,Li Z,Ji X,et al.Face R-CNN[J].2017
[2]Girshick R.Fast R-CNN[J].Computer Science,2015
[3]Ren S,He K,Girshick R,et al.Faster R-CNN:towards real-time object detection with region proposal networks[C]//International Conference on Neural Information Processing Systems,2015
[4]Liu W,Anguelov D,Erhan D,et al.SSD:Single Shot MultiBox Detector[C]//European Conference on Computer Vision.Springer International Publishing,2016:21-37
[5]Redmon J,Divvala S,Girshick R,et al.You Only Look Once:Unified,Real-Time Object Detection[C]//Computer Vision&Pattern Recognition,2016
[6]Redmon J,Farhadi A.YOLO9000:Better,Faster,Stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR),2016:6517-6525
[7]Redmon J,Farhadi A.YOLOv3:An Incremental Improvement[J].2018
[8]Wang H,Li Z,Ji X,et al.Face R-CNN[J].2017