Research And Application Of Human Posture Estimation For Home Service Robot

With the rapid development of robot and information technique,computer vision has been paid more and more attention in the field of home service robot.The human action is an important way of human-computer interaction.In addition,in order to enhance the intelligence of the home service robot,it is necessary for the robot be able to execute different actions according to the different states of the human for the same instruction,and the human action is a direct and effective condition for judging the human state.Therefore,it is an essential requirement to endow home service robot with the ability of human action recognition.However,there are huge problems in the application of human action recognition algorithm to home service robot.Human action consists of two dimensions:space and time.The robot always takes a few seconds after receive instructions to estimate human action,the delay could destroy the human-computer interaction.Due to the diversity of actions and perspectives,there is no algorithm to recognize all human actions in home environment.Because of the complexity of human action and recognition algorithm,higher computational power resources are needed,and the home service robot can't meet the requirements.In this paper,the same human posture from different actions are summarized.Additionally,the human-computer interaction is realized by recognizing static human posture.The contents of this paper are as follows:(1)The classification of human posture is proposed and the dataset is established.Based on the summary and analysis of human action in family environment,the human posture categories are proposed.The human posture dataset are established with respect categories.The labeling work also are completed according to the size and form of public datasets in order to increase the authority of dataset.(2)The scheme of human posture estimation is proposed.Based on the analysis of the human action recognition algorithm,the tradition algorithm and the deep learning algorithm are studied.The configuration requirements need to be reduced and the running speed of algorithm requires to be improved for adapting to the application of home service robot.Consequently,a scheme of human posture estimation is proposed,and the training and testing are carried out on the human posture dataset.(3)The effect of human posture recognition with small occlusion is improved.For the human posture with small occlusion,the stage method is used to estimate the human posture.Firstly,the key points of human body are recognized in the image by the key point recognition algorithm,and the coordinates of the key point are output.Then,the machine learning algorithm is used to classify the human posture according to the coordinates of the key point aiming at the improvement of the recognition effect.