The Research Of Face Recognition Based On UAV Platform

Recently,the research of face recognition based on UAV platform is gradually emerging.Many companies,such as DJI and Jingwei Hangtai,have started to implement some face recognition systems based on UAV.However,these systems have not achieved very good performance,just considering face recognition as a conceptual function.The research of face recognition based on UAV still has a lot of problems to be solved.How to further improve the accuracy of face recognition based on UAV? How far the horizontal distance and vertical height distance between human face and UAV to implement face recognition is suitable? How to solve the picture distortion phenomenon on the UAV caused by the impact of the external environment and Wide angle camera? How to accurately perform face recognition when people in different states(such as people has different hairstyles,wearing glasses or not)with very few data samples? These are all worth studying in depth.These questions are the questions to be solved in this paper.The meaning of the research in this paper is to solve these questions.The main work of this paper are:1)Establish a database of face images captured by the UAV.The UAV platform collected 21 people's face images at different heights and distances.2)Research face recognition algorithm on the UAV platform.According to the needs of this project,we choose multiclass logistic regression algorithm as face recognition algorithm for it calculates fastly and is more suitable for embedded system.Firstly,based on the experiment results on multiple public databases(ORL,AR,Muti-PIE),the multiclass logistic regression algorithm has good performance of face recognition.Then,we compare the face recognition experiment results by using two classical face recognition algorithms(Fisher Face,LBPH)and multiclass logistic regression algorithm based on the self-built UAV face database.We find that multiclass logistic regression algorithm has better face recognition performance on the UAV platform.Study the picture distortion problem captured by the UAV.We study its causes include dark light and exposure light phenomenon,jitter fuzzy phenomenon and the distorted phenomenon caused by wide angle camera.Then we give the image feature recovery method respectively.At last,the image feature recovery methods areused in real-time UAV face recognition system,which improves the accuracy of real-time face recognition from 73.21% to 80.43%.The results prove feature recovery methods are valid.3)Use a data augmentation method to improve face recognition performance,adding different hairstyles model and glasses model to human face picture.This method can improve the accuracy of face recognition when people in different states(such as people has different hairstyles,wearing glasses or not).After implementing the data augmentation method,the experimental results show that the accuracy of face recognition increases from 79.33% to 97.44%.4)Implement a face recognition monitoring system on UAV platform.Through the raspberry pie embedded system mounted on UAV to complete the functions of image acquisition,image feature recovery processing,face detection,face recognition and monitoring feedback and so on.An online experiment was conducted to find the target person from several people,and the average accuracy of identifying the target person reached 78%,which proves the effectiveness of the monitoring system.