Research On Nursing Robot Vision System Based On Deep Learning

Under the background of artificial intelligence,traditional robots can not meet people's requirements.The research of intelligent robots has become the focus of the majority of scholars,and the research of intelligent robots based on deep learning is the key to intelligent robotsThis thesis focuses on the study of service robot vision system based on deep learning.The main workflow is:First,the voice signal acquisition command is processed to inform the system of the next action command to be executed.Then the vision system performs the positioning analysis to obtain the coordinates of the object to be captured.The service robot arm grabs the object and completes the specified action.The main work is divided into the following sections:Part 1 Machine Vision,Part 2 Binocular Vision,Part 3 Visual Error Correction,and Part 4 Deep Learning Convolutional Neural Networks.In machine vision,the object recognition algorithm is mainly studied.Adaboost algorithm and support vector machine(SVM)are studied theoretically and analyzed in specific experiments.The binocular vision is the basic part of service robot vision system.Its main function is to establish a three-dimensional coordinate system based on the binocular camera imaging principle,and to perform voice recognition to obtain voice commands and perform three-dimensional positioning on the objects in the space.Binocular cameras have successfully replaced the traditional positioning methods of gyroscopes and laser sensors and are more accurate and convenient.The machine vision part combined with the binocular camera part is applied to the nursing robot for experimental analysis.A method of correcting visual errors based on BP neural network is proposed in the visual error correction section,which greatly reduces the visual error,ensures the accuracy of the system and successfully solves the problem of insufficient accuracy of visual positioning.Part of the Deep learning convolutional neural network is to optimize the reverse transmission of the convolutional neural network and successfully apply it to the system,which greatly improves the system's object recognition rate.A robotic grasping method based on deep learning is proposed.All the above four parts have been verified by concrete experiments in the nurs-ing robot vision system which is based on deep learning and experimental analysis has been conducted.The combination of the four parts is the visual system of the nursing robot based on deep learning.