The Human Detection,Tracking And Identification Authentication System Embedded In A Home Security Robot

Security is an essential part for the development of family and society.With the rapid improvement of information technology,the crime has been more technical and subtle,which undoubtedly poses a great threat to our safety and property security.Therefore,it is important to strengthen and evolve the modern security technology.With the development of science and technology,intelligent robots and its related production has greatly attracted the world's attention,especially for home security robots,which will enter the normal family to provide full range of security monitoring servicesIn this paper,we developed a human detection,tracking and identification authentication system embedded in home security robot.Specific principle of the system are described as follows: First,the system is divided into two parts,which are the mobile robots client part and the server part respectively,according to the whole function of the system.Meanwhile,we also introduced the mobile robot path planning method and server-client video transmission method.Then we utilize Fast R-CNN deep neural network for human body detection through the camera.In order to keep the target in the view of the camera,we use the keyhole imaging model and Camshift algorithm for calculating the motion of the mobile robot and angle of the camera,which could make the robot follows the target human body.Finally,the shape index feature combined joint cascade model is involved for human face detection,and the Bayesian model is involved for human face recognition.The main work of this paper include:(1).A security oriented robot system is developed.It has two parts,which are the mobile robot client platform for video acquisition and indoor inspection,and the server platform for human detection,tracking and recognition.The host and slave computer uses Raspberry Pi and Arduino microcontroller respectively,and exchange video information through WiFi.We first exploit genetic algorithm for global path planning,and then Q-Leaning algorithm for obstacle avoidance and robots movements.(2).The Deep Neural Networks(DNNs)based Fast R-CNN model and INRIA datasets trained DNN model are used for human body detection.Indicated by the experiment results,the Fast R-CNN model can detect human body in indoor scenario with robust performance,and outperforms other human body detection algorithms.(3).The camera's angle offset and mobile robot's motion offset are calculated by keyhole imaging model to keep the human target in the camera's view when the target is tractable.We utilize the mono camera demarcation methods to obtain the intrinsic parameters of the camera to diminish the distortion.Then we revise the keyhole imaging model by the tracking results of Camshift.When the target is sheltered,the Kalman filter algorithm is adopted to predict the motion trail of the target.(4).The shape index feature based human face detection and recognition methods are introduced to identify the authentication.The shape index feature is to extract the characteristics of the fixed distance range with the center point as reference,which only requires few addition,multiplication and integer calculation for fast performance.In this mobile robot,the system combines shape index feature and joint cascade model together for human face detection,then combines shape index feature and joint Bayesian model for human face recognition.