Research On Face Detection Algorithm Based On Deep Learning

As an important branch of computer vision,face detection is a crucial step in many face-related technologies and applications,such as face tracking,expression recognition,face recognition,age estimation,etc.Deep learning technology has promoted the precision of face detection to a new stage.It is widely applied in many fields such as security technology,identity recognition,information security,unmanned sales,intelligent monitoring,etc.,which have a broad space for development.Scholars and R&D personnel at home and abroad have paid more attention on face detection.Face detection still has many challenges in practical application.Some conditions such as illumination,resolution,partial occlusion,expression and posture,etc.,which will seriously affect the detection accuracy and speed.This paper studies the problem of the algorithm's robustness and real-time detection in complex environments.Firstly,the status of face detection technology at home and abroad is described.Combining with deep learning technology,the development history of face detection technology based on traditional machine learning and deep convolutional networks is analyzed in detail.Secondly,this work proposes an end-to-end cascade convolutional neural network.Then use the three-level cascade detector to perform the classification task and the bounding box regression task on pyramided pictures.The algorithm can maintain high accuracy in real-time detection while it is deployed on CPU.Thirdly,to solve occlusion,blurring and tiny faces and other issues,this work proposes a three-category face detection algorithm with context information based on region proposal network and apply sliding window to the training method.Extensive experiments on two challenging face detection benchmarks,FDDB and WIDER FACE,demonstrate the effectiveness of our method.Finally,based on the cascade convolutional neural network face detection algorithm,a face detection platform is built to perform face detection on daily photographs,scenes,etc.,which verifies the effectiveness and real-time performance of the algorithm.