Building Multi-sensor Information Fusion Map Based On Indoor Environment

With the rapid development of the mobile robot industry,the research on mobile robot platform has become a hot topic.Among them,Simultaneous Localization and Mapping are the visual presentations of the robot’s perception of the surrounding environment and are also the key to realizing the automatic navigation task of the robot platform.The speed and accuracy of robot navigation tasks are determined by map design.At present,the information source of map creation has gradually changed from singlemode information to multi-mode information.Compared with a single sensor,a map based on multiple sensor information is of greater research value because it can combine the advantages of different sensors.How to use multi-sensor information fusion to improve the speed and accuracy of map creation is an important research problem in the field of robot navigation.This paper focuses on the creation of a mobile robot multi-sensor fusion map,mainly involving the following two directions:(1)Given the limitations of map creation using a single sensor,a multi-sensor data fusion grid map creation method based on the current location storage of the robot platform is proposed.In the map creation stage,the data obtained by various sensors are processed and then associated with the coordinates of the robot platform and stored.During storage,CNN-Autoencoder is used to reduce the dimension of the collected image data to reduce the storage space.After eliminating redundant data,2D Li DAR data collected between different frames are matched and optimized by point cloud matching algorithm,to obtain relatively accurate map contour information.Compared with the single 2D Li DAR map creation method,this map creation method retains more map information on the premise of ensuring the speed of map creation and reducing memory consumption.Experimental results show that compared with the RGB-D image point cloud mapping method,the speed of the map creation method proposed in this paper is improved by 31% and the space occupation of the map file is reduced by 47%.(2)A loop detection method based on image information and pose information is proposed according to the errors caused by environment and hardware in the map creation stage of mobile robots.Firstly,the K-Nearest Neighbor classification algorithm is used to match the image information after dimensionality reduction,and the image information that is close to the sample image in the storage is selected.Then,the Triplet network with Res Net as the main stem network is used to accurately match the filtered image,so as to judge whether the image is in the same environment.Finally,the accuracy of loop detection is improved by comparing robot pose information.Experimental results show that the proposed method improves the matching speed of image similarity by 48% and the accuracy of loop detection by 8% compared with the traditional method.