Research On Auto-following System Of Multi-sensor Fusion

Autonomous following robot is an important part of achieving human-machine integration environment,and its tracking and location of the target person is an important prerequisite for achieving following.Robust following planning and control provide a guarantee for the continuous following.This project aims to design a pedestrian following robot with multisensor fusion for unknown environments.The main work and contributions are as follows:1.Overall platform of the following robot is constructed.The target pedestrian tracking and positioning system is built based on camera,lidar and ultra-wideband following module,and SCOUT,which integrates the information of inertial measurement unit(IMU),is used as the mobile platform of the robot.Based on the ROS framework,the overall structure of the following system is proposed.2.Design of pedestrian tracking and positioning system for following robot targets.Using deep learning methods YOLOv7 and Deep SORT to track pedestrians within the robot’s field of view,fusing lidar point cloud data to extract corresponding point clouds for pedestrians and achieving pedestrian positioning in real space.Determine and locate the target pedestrian based on the information of the ultra-wideband following module.Additionally,according to the data characteristics of the sensor,a follow behavior state machine including follow behavior,transition behavior and recovery behavior is proposed,which ensures the robustness of the follow by switching the behavior during the follow process for different situations.3.Design of following robot planning and control system.Based on the target location information and the surrounding obstacle information,an improved dynamic window algorithm(MDWA)is used to implement the following planning and control that supports the following behavior state machine.That is,follow-up behavior in normal state,recovery behavior is used to retrieve the target when the target is lost,and transition behavior is used to concatenate follow-up and recovery behavior to ensure the robustness and continuity of follow-up.For the system proposed above,this paper conducts a lot of experimental verification in real scenes.The experimental results show that the robot can switch its behavior by following the state machine,which guarantees both obstacle avoidance and pedestrian following after determining the target.It also keeps track of the target when there are multiple pedestrian disturbances in the following scene.When the target pedestrian changes due to light conditions or turns are lost,the robot can still retrieve the following pedestrian by switching behavior.Particularly for the loss of pedestrian turn scenes,several experiments were performed,and the success rate of finding the target after turning was 81.4%.In addition,it is compared with other follow-up methods to show the advantages of this method.