Research On Semantic Mapping Based On FCHarDNet

Environment perception is an important part of robot intelligence.To better understand their surroundings,robots need to know not only the shape of objects in the real world,but also their semantics.The moving position of dynamic objects in the environment may affect the accuracy of SLAM algorithm.This is the case because SLAM algorithms typically assume that only the robot is moving in the surrounding environment and that all other objects are stationary,while using distance data to generate a map and estimate where objects are in space.If the surrounding environment includes dynamic objects,they may remain in the generated map or be identified as stationary objects,which can lead to errors in location estimates.To solve this problem,DF-SLAM algorithm is proposed in this thesis.In the research and development of SLAM,the camera is usually used to obtain environmental information to build a three-dimensional spatial map,and then the extracted image feature information is sent to the world coordinates through feature extraction and descriptor matching,and three-dimensional point cloud map is presented.There are obvious problems when intelligent robots use point cloud maps to complete navigation.Point cloud images have a large scale and store a lot of unnecessary details,so they take up a considerable amount of storage space,resulting in low storage efficiency.Octo Map saves a lot of space compared to point clouds.Each voxel of Octo Map updates the occupancy rate from different measurements in a probabilistic way.However,Octo Map lacks high-level semantic information,and no semantic color and semantic confidence information are stored in voxels.To solve these problems,a semantic Octo Map mapping method based on FCHar DNet is proposed.(1)DF-SLAM is a SLAM algorithm that combines semantic segmentation and deep learning techniques to achieve indoor and small-scale outdoor navigation and scene understanding.The core idea of DF-SLAM algorithm is to combine semantic segmentation with traditional visual features.This algorithm can detect,track and recognize objects in dynamic environment,and at the same time carry out camera positioning and map construction,so as to provide more accurate and robust navigation services in the scene.The DF-SLAM algorithm was tested on multiple data sets and compared with other existing SLAM algorithms.The experimental results show that the DF-SLAM algorithm can provide more accurate and robust localization and map construction capabilities in dynamic environments,and can recognize objects in the environment semantically,providing more information for indoor navigation and scene understanding.(2)DF-SLAM is a SLAM algorithm that combines semantic segmentation and deep learning techniques to achieve indoor and small-scale outdoor navigation and scene understanding.The core idea of DF-SLAM algorithm is to combine semantic segmentation with traditional visual features.This algorithm can detect,track and recognize objects in dynamic environment,and at the same time carry out camera positioning and map construction,so as to provide more accurate and robust navigation services in the scene.The DF-SLAM algorithm was tested on multiple data sets and compared with other existing SLAM algorithms.The experimental results show that the DF-SLAM algorithm can provide more accurate and robust localization and map construction capabilities in dynamic environments,and can recognize objects in the environment semantically,providing more information for indoor navigation and scene understanding.