Improvement Of Face Detection MTCNN Algorithm In Classroom Environment

With the rapid development of deep learning technology,face detection is widely used in intelligent transportation,financial security and other fields due to its important academic value and application value.In view of the complex and variability of natural scenes and face itself,face detection poses a huge challenge in complex scenes.At present,the face detection algorithm based on deep neural network in deep learning has achieved better results than traditional algorithms in both academia and industry,and has been highly valued.Compared with the R-CNN series general target detection method,the face detection MTCNN(Multi-Task Convolutional Neural Network)algorithm has the characteristics of fast face detection and small space occupancy.The face image in the classroom environment is affected by factors such as illumination,face orientation and occlusion,which makes the face detection accuracy based on MTCNN low and the false detection rate is large.Based on the fact that MTCNN is not ideal for non-positive face detection,this thesis deeply studies and improves the face detection MTCNN algorithm in classroom environment.The main research work and results include:(1)MTCNN algorithm is introduced into face detection in classroom environment,and face detection is compared between standard face data and classroom environment face data.The detection effect and false detection are analyzed concretely.It is concluded that face orientation in classroom environment is one of the factors affecting the accuracy of face detection.(2)Based on the limitation of the number of training samples in the classroom environment,the MTCNN network model parameters are migrated and learned,which reduces the training time of the model.The model after parameter fine-tuning reduces the repeated prediction frame of the face.(3)The face feature point alignment module is adjusted to improve the MTCNN network model structure,so that the face detection of the model in the classroom environment face test set does not depend on the key features of the face feature,which improves the detection time efficiency and the face detection accuracy rate of the algorithm.(4)Aiming at the misdetection of MTCNN in the classroom environment,according to the model to detect the score given by the face on the face image,a false detection discriminant based on face similarity is proposed,which greatly improves the false detection.Finally,through the comparison of multiple experimental results,the effectiveness of the improved algorithm in this thesis was preliminarily proved,and the improved strategy significantly improved the accuracy of face detection in the classroom environment.