Research Of Robot Semantic SLAM

With the development of emerging artificial intelligence technology,robots inevitably entered the stage of intelligence.How to make robots perceive and observe the world like humans has always been a topic of computer science.Simultaneous localization and mapping(SLAM),which mainly solves the problem of "Where am I?" And "What is my surroundings?" In robots,is the key to robots to achieve autonomous intelligence.In recent years,extensive research and application of Object Detection and semantic segmentation based on deep learning have made it possible to obtain very accurate semantic information.Further,Combining semantic information of objects in the environment into a SLAM system is one of the hottest research fields of SLAM,which is called Semantic SLAM.In response to the needs of future indoor robots for advanced semantic information,the research in this paper mainly has two aspects: First,research on object detection algorithms that meet the requirements of SLAM systems.It is not only satisfy the real-time performance of the SLAM system,but provide accurate object detection results for the SLAM system.The second is how to apply object detection technology to visual SLAM to improve the robustness and speed of the system.To this end,the work of this article is divided into as following:(1)Research on improved object detection algorithms.Aiming at the problems of current object detection algorithms,based on the considerations applied to SLAM systems,this paper designs an improved YOLOv3 algorithm that compensates for the loss of information caused by convolution operations.While meeting the real-time requirements of SLAM systems,it can give SLAM an accurate semantic information.(2)The application of the improved object detection algorithm proposed in this paper to SLAM front-end visual odometer.In visual SLAM based on feature points,for pose estimation and feature point matching in visual odometers,this paper proposes a pose detection calculation method based on object detection and feature point detection based on object detection in a matching algorithm.The method that using the position information and semantic information of object has greatly improved the operating speed of the SLAM system.(3)Study the application of the improved target detection algorithm proposed in this paper to SLAM closed-loop detection.A closed-loop detection method using objects in the surrounding environment and their relative positional relationship was designed and applied to traditional visual SLAM.