Research On Visual Localization And Navigation System Of Indoor Mobile Robot

With the development of the era,the indoor mobile robots have many applications scenarios such as restaurant delivery,airport cleaning,bank services,etc.However,due to the immatureness of current software and hardware technologies,indoor robots haven't been widely used.In localization and navigation technology,the use of visual sensors has the advantage of obtaining richer environmental information than laser radar,so as to build a more accurate environment map.But its localization robustness needs to be improved and the way of map building needs to be changed.Based on the theory of the robot localization and navigation system,this paper optimizes the localization algorithm and map building method and adopts the obbec RGB-D camera to implement robot localization and navigation function.The main work of this paper is as follows:(1)Research on the principle of positioning and mapping system.The whole framework of visual positioning navigation system is analyzed,and the probability problems and optimization method in the process of simultaneous localization and mapping are mathematically derived.(2)Image data processing.The distortion of color and depth camera lens in imaging process is analyzed and calibrated.Aiming at the problem of noise depth data,this paper proposes a multi frame fusion Gauss smooth model based on the error model of the existing depth sensor which effectively removes the depth image's noise and fills partical missing depth data.(3)Research and optimization the location algorithm.The feature point localization method and image gray information location method are studied to analyze the advantages and disadvantages of the two methods.A direct method is proposed,which takes semi dense photometric error and dense geometric error as the optimal target.On this basis,this paper designs an adaptive motion estimation method which effectively improves the robustness of robot localization in low texture environment and its overall environmental adaptability.(4)Research on the method of map building.Aiming at the problems that 3D map building is not easy to update and 2D fake laser map is inaccurate when RGB-D camera mapping,this paper proposes a method to construct 2D grid map with 3D information fusion which makes the map more accurate and easier for navigation.(5)Design the robot navigation system.The A* algorithm is used to plan the global path,and the VFH algorithm is equipped with an infrared ranging sensor and an ultrasonic sensor to develop the dynamic obstacle avoidance function of the robot.(6)Build hardware and software platforms for experimental verification.Firstly,the improved algorithm is tested in the TUM standard dataset.Secondly,the robot hardware platform required for this experiment is built,and the localization and navigation software system developed in this paper is transplanted.Finally,in the indoor space of more than 100 square meters,the localization accuracy,tracking robustness,map accuracy,and navigation process of the robot are actually tested.