Active Mapping And Obstacle Avoidance In Navigation For Wheeled Mobile Robots Based On Indoor Visual Localization

Autonomous mobile robot is a high-tech integrated product involving mechanical design,circuit analysis,sensor technology and software algorithms.With the rapid development of artificial intelligence technology and the active promotion of national policies,autonomous robots have been widely used in manufacturing,medical and service industries.The importance of highly customized artificial intelligence algorithms that relate to the level of intelligence of mobile robots is self-evident.Simultaneous Localization and Mapping(SLAM)plays an vital role in intelligent algorithms for autonomous mobile robots.It integrates the sensor information through algorithm fusion to realize the real-time localization of the robot and the mapping of the environment.According to the type of sensor,SLAM algorithm can be divided into laser SLAM and visual SLAM(vSLAM).Compared with the laser solution,the visual solution has the advantages of low hardware cost,rich information dimensions,and easy installation of cameras.This makes the vSLAM method to be more potential for environmental perception.The traditional vSLAM method is a passive sensing procedure that requires human participation,which limites the application scenarios,so it can't cope with some completely unmanned scenarios such as unmanned inspections,indoor rescue and unmanned cleaning.Therefore,the active vSLAM based on autonomous exploration strategy came into being.It can realize the localiztion and active mapping unknown environment.Focusing on the enhancing of the intelligence of autonomous mobile robots with active vSLAM and the perceptual advantages of RGB-D cameras.This thesis investigates the active mapping and navigation obstacle avoidance of autonomous mobile robots based on indoor RGB-D visual localization.The main research contents and innovations are as follows:(1)A SLAM method based on graph optimization for combing wheel odometer with vision is proposed.The vSLAM converts the input visual information into localization and map information.Robust and accurate localization output is a prerequisite to ensure the continuous and stable running of the system,and it is the main factor affecting the accuracy of mapping process.Comared with the existing method of fused wheel odometer information,this theis proposes an online estimation method of camera installation pose based on graph optimization,which is used in the fusion SLAM method of wheel odom-eter and vision,which overcomes requirements for the pre-calibration results of the camera installation posture.By extracting the ground plane and matching feature points,use the graph optimization method to construct plane constraints and matching point constraints to estimate the camera's installation pose that will be uesd in the odometer-constrain construction process of the back-end pose optimization,and finally achieving the purpose of fusing wheel odometer information and improve the visual localization performance.Field data are adopted to verify the effectiveness of the proposed method.(2)An active exploration navigation path generating method is put forward based on image morphology for object detour.The above non-absolute visual locating method will inevitably lead to cumulative error under long-term running,while the loop-closure optimization is an effective way to eliminate cumulative errors.And it usually be carried out with the subjective knowledge of the operator in traditional passive vision SLAM method.The existing active SLAM generate loop-closure require revisiting a loop-closure point by passing through the known area.The revisiting type can't meet the "loop" requirement.Besides,inconsistent image angle is not conducive to triggering loop-closure detection.Thus,this thesis proposes an active exploration strategy for object detour based on image morphology processing.By approaching the boundary between the known map and the unknown map,orbiting the instance objects in the environment,it can improve the map coverage while active generating loop-closure optimization.In the end,the effectiveness of the strategy is verified in a simulation environment.(3)A modified velocity information layered-costmap is proposed.The navigation algorithm's performance determines the robot's control autonomy and operational safety.While the methods based on layered costmap have problems with dynamic obstacle avoidance,and the existing improved methods have problems of poor adaptability and difficulty in adjusting parameters.Thus the paper proposes an improved layered-costmap method that integrates dynamic obstacle velocity information.By dynamic obstacle extraction and tracking,modifying the core cost calculation algorithm,an improved costmap is generated for subsequent global and local path planning.In addition to the existing navigation framework,a dynamic path re-planning mechanism has been added to improve the real-time accuracy of avoidance paths.In the end,the improvement of navigation and obstacle avoidance performance is verified by experiments.