| In recent years,face detection technology has become one of the hottest research topics.With the maturation of face detection technology,its use in society is also seen everywhere.In real life,banks,shopping malls,customs,roads,The extensive use of technology such as station security inspections and other security monitoring has increased the demand for face detection,not only for the accuracy of detection but also for the speed of detection,as far as the current status of face detection is concerned.Due to the low efficiency of multi-view face detection based on complex background,the high rate of missed detection has become a difficult issue in current research.This article mainly focuses on the research on the problems of multi-view face detection.This paper proposes a multi-view face detection algorithm based on convolutional network.This algorithm uses popular deep learning to replace the traditional face detection algorithm in recent years,and fully uses the advantages of deep learning to improve the detection accuracy.The experimental results under the framework of Caffe show that the proposed method has superior performance in terms of efficiency and precision,and it also has certain robustness to skin color and illumination.The main work of this article is as follows:(1)For the problem of face multi-view determination in complex background,the attitude algorithm is used to solve the multi-view problem.This paper uses the simplified model based on the Candide-3 standard model,which is simplified from the 113 vertices in the standard model to the face pose.The 88 vertexes innovatively increase the labeling of the key feature points of the center of the left and right pupils,the tip of the nose,and the center of the mouth.Eventually,through the projection of the position of the feature points on the three-dimensional space,the feature triangles at the position of the plane are obtained,and finally the deflection angle is obtained through mathematical knowledge..By comparing the AFLW standard multi-pose database with other mainstream gesture algorithms,the effectiveness of the Candide-3 model method adopted in this paper is finally verified.(2)Aiming at the traditional face detection method,as one of the mainstream convolutional neural networks,its advantage is highly sought after.Based on this,this paper proposes a multi-view face detection algorithm based on convolutional neural network.AFLW multifaceted face database obtains the original positive and negative sample datasets.After the normalization process of sample images and the construction of Candide-3 model,the Network In Network(NIN)network structure model is used to train a deep face classifier.As a result,face detection is finally performed.(3)Aiming at the problem of finally obtaining multiple face frames through the detection method of this paper,this paper proposes a non-maximum suppression(NMS)algorithm to select the optimal face markup frame for each person.The optimal solution of the iterative search locality in the face tagging frame is finally determined as the optimal face detection window,and the cross-over and poorly labeled face frame is eliminated. |
