| The increasing application of mobile robot technology in various domains has been propelled by the continuous advancements in hardware and software technologies.The market for mobile robots,including industrial robots,consumer robots,commercial robots,unmanned vehicles,and drones,is experiencing rapid growth.However,traditional visual Simultaneous Localization and Mapping(SLAM)methods,such as vision-based SLAM and 2D laser mapping,fail to meet the evolving demands of mobile robot development.To address this challenge,this study investigates an enhanced approach to the development of a mobile robot RGB-D visual SLAM system based on YOLOv5 dynamic detection.Additionally,the fusion of single-line Li DAR and RGB-D camera information is explored for improved planning capabilities.Most visual SLAM systems assume scene rigidity,neglecting the presence of dynamic objects in the distribution of feature points.This limitation results in decreased accuracy in camera pose estimation,consequently affecting the localization and mapping accuracy of the camera coordinates.To overcome this limitation,we propose an RGB-D visual SLAM system based on the YOLOv5 detection network.By incorporating dynamic object detection,we achieve more precise camera motion pose transformations.This is achieved by calculating the relative poses of dynamic objects in consecutive frames and assigning confidence to dynamic relative poses.Subsequently,the compensation of the camera’s relative pose matrix enhances the accuracy of camera pose estimation.To fully leverage the RGB-D camera’s capabilities,a thread for generating dense point cloud maps is integrated into the system.This facilitates efficient robot navigation and localization,while minimizing the impact of dynamic objects on camera pose estimation,ultimately providing a practical dense map.To enhance the mobile robot’s environmental perception and ensure safe and stable operations,the study integrates single-line 2D Li DAR information and fuses the grid map obtained through the Gmapping algorithm with the point cloud map stored by the improved visual SLAM system.By incorporating the point cloud map generated by visual SLAM into the grid map and integrating three-dimensional features,a fusion map with improved accuracy and information richness is created.This fusion map enables accurate and secure path planning and navigation.The effectiveness of fusing single-line Li DAR and RGB-D sensor information is validated through practical scenarios,reducing the need for frequent adjustments to the inflation radius and enhancing path planning efficiency and mobile robot perception capabilities.To address the high-cost requirements associated with incorporating large-scale deep networks into VSLAM systems,which hinder the widespread adoption of mobile robots,this study presents a cost-effective solution.A experimental platform based on Jetson Nano mobile robots is designed,utilizing the Robot Operating System(ROS)on a Linux platform for implementing the improved system verification,map information fusion,and path planning modules.The results demonstrate that the proposed system achieves high localization accuracy and disturbance resistance in dynamic scenarios.Moreover,the utilization of the fusion map enables excellent path planning outcomes,ultimately enhancing the autonomous operation capabilities of mobile robots.This study provides a concise and convenient approach for the fusion of map information in mobile robots,expanding the application of VSLAM in different scenarios and improving the perception capabilities of mobile robots in terms of localization and mapping. |
PDF Full Text Request