Research On Key Technology Of Face Recognition Based On Unconstrained Conditions

Face recognition is a non-contact human-computer interaction technology.Compared with other human biological features,face features have the characteristics of security and stability that cannot be copied.Therefore,human-computer interaction technology based on face recognition has important theoretical research value and practical value for the development of artificial intelligence.The overall structure of the face recognition system is designed,and research of face detection is also completed at first.Aiming at the poor robustness of face detection under unconstrained conditions such as illumination and noise,an improved MTCNN face detection algorithm is proposed in this thesis.Firstly,bicubic interpolation is used to preprocess the original face image.Secondly,non-maximum suppression is adopted to remove the wrong face detection box.Moreover,bounding box regression is applied to fine tuning the face detection box.Finally,an accurate face detection box and four face feature points are obtained.Experimental results show that the improved MTCNN can not only detect the face accurately but also locate the face feature points correctly under unconstrained conditions,which lays the foundation for the face feature extraction.In the stage of face feature extraction,aiming at the problems that the loss of face feature information under unconstrained conditions such as illumination and noise,and the single local feature extraction method will affect the result of face recognition,a novel fusion method of improved adaptive local ternary pattern(IALTP)and two-dimensional two-directional principal component analysis((2D)~2PCA)for face feature extraction is proposed in this thesis.Firstly,the IALTP algorithm is used to extract the face texture feature information of the four face feature points regions as the local features.Then,the global feature of face can be obtained by(2D)~2PCA.Finally,the local features and global feature are merged to obtain the IALTP+features.Experimental results that this method can extract face texture information accurately,and obtain a higher face recognition rate under unconstrained conditions.In the stage of face feature classification,the eXtreme Gradient Boosting(XGboost)is adopted to classify.By introducing the two order Taylor Expansion to calculate the loss function,the regular term is added to the loss function in order to prevent data fitting,and the CPU multithread parallel computer system can speed up the calculation,which have played an important role in improving the accuracy of face recognition in real-time.The experimental results show that the average recognition rate of face recognition is96.2%,and the average recognition time is 75ms under non-constraint conditions.Finally,the platform verification of the face recognition system based on mobile robot is completed in this thesis.The result of the face recognition is transformed into an instruction to control the movement of the robot.The accuracy and reliability of the algorithm is verified by a large number of experimental tests.It is proved that this method can achieve the robot motion control based on face recognition.