Research On Object Search Method Of Family Service Robot Based On Visual Attention Mechanism

With the development of artificial intelligence technology and the aggravation of population aging problem in China,the research on indoor service robot has become a hot spot in the field of automatic control and computer technology.In order to make service robots have high efficient and precise visual perception ability and improve the service level of the robot,so the research of robot vision algorithm has become the research focus in the field of robot.With the rapid development of image processing and computer vision technology,the algorithms used in the field of robot vision are endless.But the service robots work in complex and unstructured family environment,so it is a challenging task to realize service robot's recognition and understading of environment with complex background disturbance and occlusion.Therefore,it is of great significance to study a set of visual identification methods with wide application scope,high accuracy and high real-time performance.The target search method of service robot in the home environment is studied in this paper,and a target search system based on visual attention mechanism is proposed.In this paper,the application of the human vision selective attention mechanism in robot vision algorithm enables the robot to catch useful interset area like human quickly.It reduces the retrieval and identification for unrelated areas and the service efficiency of service robots is improved greatly.The proposed robot target search system mainly includes three parts:target area preselection based on visual attention mechanism,service robot path planning based on improved genetic algorithm,target precisionidentification and matching based on WSMTL model.This article mainly completed the following work:Firstly,the VOCUS preselection system based on visual attention mechanism was studied.The system combines the Bottom-up and Top-down visual attention mechanism,the interset area related to target in the scene is detected cooperatively as the preselected area of target.So the filtering of visual information reduces the irrelevant area detection.To a certain extent,it improves the efficiency of the service robot target search.Secondly,the service robot path planning method in the home environment is studied.We propoed a path planning method based on improved genetic algorithm,it introduced the concept of visual space so it make sure all the indivisuals are produced,also the convergence rate of the algorithm is accelerated.The genetic operator and mutation operation were improved so the diversity of quasigroup was increased.The path palnning model can provide the robot with the shortest and unobstructed path from the initial position to the target,so that the robot can realize the target recognition of near distance.Thirdly,the precision recognition and matching algorithm of robot target is studied,and a target recognition model based on weak supervision and template learning is proposed(WSMTL).A simple and effective template matching algorithm is used for target recognition task,and the template is trained from training data.A new framework which can suppress the impact of noise and unbalanced training data is designed,and it can automatically select "clean" images for template learning and recognition.Through WSMTL model,the robot can identify and match the target near distance and improve the accuracy of target recognition.