Outdoor Simultaneous Location And Mapping For Unmanned Ground Vehicles With Binocular Camera

With the fast development of mobile robotics, the research of unmanned ground vehicles has become the focus of the intelligent vehicles’field. Being able to build a map of the environment and to simultaneously localize within this map is an essential skill for mobile robots navigating in unknown environments. This so-called simultaneous localization and mapping (SLAM) problem has been one of the most popular research topics in mobile robotics for the last two decades. SLAM-oriented research in a wide range of outdoor environments, graph-based approach is the current hot topic. In a graph, nodes correspond to the poses of the robot at different points in time and edges represent constraints between the poses. There are three steps in graph-based SLAM task. They are graph construction, loop closure detection and optimization. We explored three of them in campus environments.A stereo image matching algorithm is deployed to perform consistent pose estimation for UGVs so that an initial pose-graph model is constructed for the UGV’s SLAM task. This uncorrected pose-graph map should be optimized with the spatial configuration constraint generated by the loop closure detection algorithm during the UGV’s navigation. A loop closure detection algorithm based on ORB feature matching and Bag-of-Words model is utilized in our work, which can provide the constraints of temporal consistency and of geometrical consistency to improve the accuracy of loop closure. The back-end implementation for graph-based SLAM is used the Gaussian-Newton optimization method and the sparse characteristics of the system information matrix is fully utilized in the iterative procedure. To improve the computing speed, sparse Cholesky decomposition was used to solve matrix equations. Finally a revised maximum likelihood topology of the UGV was acquired with the minimum system error.To test the validity and robust performance of the proposed approach, experiments were conducted on our self-developed UGV in DUT campus. Results show that the loop closure detection algorithm we proposed can accurately detect the area of closed-loop in real time during UGVs’ travelling, thus can provide effective constraints for back-end optimization.