Research On Autonomous Localization And 3D Semantic Mapping Algorithm For Mobile Robot

With the opening of the curtain in the era of intelligence,robot intelligence has become a hot topic for scholars at home and abroad.Autonomous Navigation of mobile robot is the key technology of intelligent development of robot.The core of this technology is to realize the real-time positioning of robot itself,as well as the perception and understanding of external environment.The technology of simultaneous localization and mapping(SLAM)provides a theoretical basis for the real-time positioning and environmental perception of mobile robots,while the technology of visual-based SLAM can make mobile robots have the ability to understand the environment and make the intelligent development of mobile robots possible.At present,the visual SLAM technology still has some problems,such as poor robustness,unable to generate semantic map suitable for autonomous navigation.Therefore,this paper aims at these problems,combined with instance segmentation technology,to carry out in-depth research on visual SLAM.Visual SLAM is mainly used to estimate the pose of robot in real time and build incremental map by using the observation data of vision sensor.In this paper,an RGB-D camera is used as the sensor of the robot,and a visual SLAM scheme based on instance segmentation is proposed.First,while extracting feature points from the sensor output image,the improved instance segmentation network is used to perform instance segmentation on the image.Then,the target detection information of the instance segmentation is used to assist the positioning,to eliminate the feature points that are easy to cause mismatches,to narrow the area of feature matching,and the semantic information of instance segmentation is used to realize the loopback detection function.Finally,the three-dimensional semantic map is built to implement reuse maps and human-computer interaction.This paper uses TUM data set for experimental verification.The experimental results show that the proposed scheme can increase the robustness of image feature matching,speed up the feature matching speed,and improve the accuracy of mobile robot positioning.And the scheme proposed in this paper can generate accurate semantic maps to meet the needs of robots performing advanced tasks.