2D Object Semantic Map Construction Oriented To Efficient And Safe Navigation In Home Environment

To complete complex service tasks and achieve intelligent and long-term autonomous operation in home environments,service robots need efficient and safe navigation,which can effectively plan paths and avoid collisions with any part of obstacles in order.Maps can provide prior information about obstacles in the environment and are an important basis for autonomous navigation.As an environment map that represents both obstacle distribution and object attribute information,object semantic map can provide more rich prior guidance for efficient and safe navigation of service robots in unstructured home environment.At present,there are some problems in the obj ect semantic map construction,such as weak correlation between metric maps and semantic information,the inadequate ability to handle semantic uncertainty,and the insufficient support for navigation application.For these problems,this paper proposes a 2D object semantic map construction system oriented to efficient and safe navigation in home environment to meet the complex navigation requirements of service robots.The system can enhance the semantic location association,improve the accuracy of semantic information and enhance the support of semantic maps for navigation.The semantic map building system uses a 2D structure map based on RGB-D camera as a metric map.The object semantic information is extracted and the semantic uncertainty is handled during mapping.On this basis,this paper closely associates obstacle location data with semantic information to form an object semantic map framework for navigation applications,which ensures that semantic maps support navigation efficiency and safety.The main tasks are as follows:A 2D structure map construction method based on an RGB-D camera and corresponding auxiliary map construction strategies are presented to represent the obstacle information more comprehensively.The construction method can accurately generate the maps reflecting the top-down contours of obstacles in the environment.In order to convert the raw data of RGB-D camera into virtual laser data and improve its compatibility with 2D SLAM algorithm,a virtual laser generation algorithm based on polar scanning is designed.The scanning process is adapted to the camera’s visual characteristics through effective projection range detection and discrete point group optimization.The auxiliary strategies proposed in this paper improve the quality of 2D structure map by eliminating bad quality observations without changing the SLAM algorithm and the generation process of observations.An object semantic extracting method based on cell grid representation and keyframe update is presented,which can recognize the semantic tags of objects from real-time visual data accurately and associate their location information.In the method,the point cloud optimization based on height filtering and the local grid optimization based on visual area ratio are used to remove the remarkable incorrect semantic data through the object 3D features and visual joint features.At the same time,in order to deal with the more random semantic uncertainty,global grid update and optimization methods based on a voting mechanism are proposed according to the statistical characteristics of the occurrence number of times of cells at different keyframes.On this basis,this paper presents an object instance division method based on distance distribution and regional constraints using the overall geometric features of semantic information,which achieves a fine partitioning of the semantic data in the global semantic cells from category to instance.A Navigation-oriented object semantic map framework is proposed,which unifies the semantic information of objects and 2D structure map into a semantic map,and implements the whole process from efficient storage to navigation application.In the framework of semantic maps,in order to represent discrete semantic location data as contiguous location areas and achieve a uniform representation and association with 2D structure maps,a semantic description rule based on convex polygon is proposed and a semantic map is stored as description files.In order to enhance the support of semantic map for navigation and improve navigation efficiency,this paper designs semantic parsing algorithm and semantic navigation interfaces,which can integrate and infer semantic data.