Pedestrian Following Method For Indoor Mobile Robot Based On Depth Image

With the rapid development of artificial intelligence,related technologies are gradually applied to people's daily lives.Products such as image processing,voice recognition,robotics,and intelligent search began to enter into all aspects of medical care,education,finance,travel,and industrial production.In the process of industrial intelligentization,robots not only replace the manual and dangerous tasks,but also perform high-precision and high-complex tasks.In the field of mobile robots,different robots are developed according to different application scenarios,including AVG in the logistics industry,robots in the service industry,and unmanned vehicles on roads.Among them,map creation,path planning and target tracking are the core technologies of mobile robots and the key to the successful application of mobile robots.Due to the complex indoor environment,it is easy for mobile robots to lose track of the target.This thesis will focus on the follow-up issues of indoor mobile robots as follows:(1)Indoor mobile robots need to obtain depth information of the target.Depth information is obtained by a laser sensor,but the price is high,and the amount of calculation is large by other methods such as binocular vision.Here,the Kinect depth sensor is used as the main sensor of the mobile robot and can simultaneously obtain depth information and color images.In order to improve the real-time performance of the mobile robot,the depth image is simulated as a laser,which not only retains the depth information but also reduces the amount of calculation.For the void and flicker problems in the depth image obtained by the Kinect depth sensor,the frequency-based method is used for filtering,and for the holes that still exist after filtering,the region growing method and the median filter are used to eliminate the holes.(2)For the indoor mobile robot following the scene,the human leg is used as the main tracking target.The feature descriptors of the target human leg are designed,and the clustering method based on multi-constraint conditions is designed for the laser data.Label the clustered targets,train the SVM classifiers,and cross-validate the trained classifiers.The position information of human legs is extracted through a trained SVM classifier.(3)A method for pedestrian tracking of mobile robot based on laser data is designed.In order to improve the tracking robustness,a Camshift target tracking algorithm that fuses the target depth information is used.Due to the limited mobility of the mobile robot and the limited field of vision of the Kinect sensor,the target is easily lost in the field of view.In view of this situation,we analyzed the target's motion information and designed a method for finding the target after losing the target.We determine the following target through the laser data feature and the Camshift histogram information,so that the mobile robot can follow the target again.(4)Turtlebot mobile robot is taken as the experimental platform,the basic knowledge of ROS robot operating system is introduced.The software structure of the target following system is designed.The effectiveness of the target following algorithm is verified through experiments,and the effect of finding the target and re-following the target after the target is lost was verified.