Research On Application Of Face Recognition Based On Deep Learning In Intelligent Welcome Robot

As the foundation and premise of the intelligent welcome robot decision module and action module,the perception module directly affects the production and application of the intelligent welcome robot and its large-scale popularity.As one of the most commonly used technical methods in the perception module of intelligent perception robot,its recognition accuracy plays a decisive role.However,traditional face recognition methods have problems such as insufficient accuracy and anti-interference,which cannot be applied to intelligent welcome robots with extremely high face recognition requirements.With the advent of the information age,the feature-based extraction method has been widely used in many aspects.Compared with the traditional method,deep learning uses a multi-layered input data set to extract nonlinear mappings from the underlying features to advanced features by simulating the human cerebral cortex.In this way,the complex feature extraction is simplified and abstracted and the human learning process has proved to be more effective.From the deep belief network proposed in 2006 to the convulsion neural network that appeared in 2012,both the theoretical research and the practice have promoted the continuous progress of deep learning.Therefore,this dissertation focuses on the face recognition in the vision of smart welcome robots.For face recognition based on deep belief network,there still exist some problems such as weak local feature learning,noise and attitude in the process of face feature learning,which make the face recognition accuracy difficult to meet the recognition requirement of intelligent welcome robot.Based on the above reasons,firstly,this paper analyzes the local two element model through theory and experiment,and finds that the face image feature extraction process has the advantages of small computation and strong anti noise ability.Therefore,the local two element pattern features of the face image are used as the input of the depth belief network to realize the depth belief network which fuses the LBP.The experimental results after adding the optimization techniques show that the method can learn the local texture detail features of face images.Compared with the single LBP method and the depth confidence network method,this method has a satisfactory recognition rate in the prediction process.At last,as deep learning changes from theoretical research to practice,more and more open source framework applications can be optimized and improved combined with business and then applied to actual systems.Based on the engineering project of the intelligent welcome robot,this paper analyzes the existing problems of the face recognition method of the depth belief network of the fusion LBP and also uses SeetaFace's open-source face recognition engine to complete a face recognition system,through the module testing and system joint debugging verifies that the method can be successfully applied to practical projects and basically achieves desired planning goals,reaching the intelligent welcome robot face recognition needs.