Design And Implementation Of Visual Localization Algorithm For Demo System Of Construction Robots

Construction robot is a kind of robot that works in unstructured environment and performs complex construction tasks by its own power and control capability.The robot must have positioning capability for motion planning and material transporting.Compared with traditional positioning methods,visual positioning has the advantages of wide application,rich information and low cost.This paper originated from a project of designing a visual positioning scheme for a demonstration system in which excavator models move autonomously on an indoor sand table.In order to implement a robust,accurate,real-time,low-cost visual positioning system that can be applied to the variable sand table terrain,this paper proposes a positioning scheme combining image feature and optical flow algorithms based on offline reconstructed visual maps.And some special improvements is made for the requirements of the project.This research includes the following aspects:Firstly,the characteristics of the project and the accuracy,speed,and robustness requirements are analyzed,and various visual positioning methods are studied.Then the visual positioning scheme is designed.According to the scheme,visual scene design and sensor selection and calibration are carried out,and the algorithm modules of visual mapping,map processing,image feature positioning and optical flow positioning are defined.Secondly,according to the requirements of the demonstration system,the method to obtain the visual map required by positioning is designed.The monocular sparse incremental 3D reconstruction based on SIFT features is adopted to extract the visual features of the scene around the sand table and calculate their positions.The visual map is then aligned to the global coordinate system and the pose between the camera coordinate system and the body coordinate system is calibrated.And the map is processed to obtain a simplified and stable visual map.Then,in order to implement the robust real-time positioning of the robot,a visual positioning algorithm combining image feature positioning and optical flow method is designed.The images feature method uses SIFT features,selects the relevant pictures in the map by criteria of word bags etc.,and uses Pn P algorithm to calculate the camera pose.The optical flow method that tracks the feature points obtained by the feature method is designed to improve the real-time performance of the algorithm.And the parallel processing of the two methods and failure recovery is designed.Finally,the above algorithm is transplanted to the experimental mobile platform,and the mapping and positioning performance is tested on the demonstration system to examine the accuracy,real-time and robustness of the algorithm.And the feasibility and improvements of the algorithm when applied to real construction robots are discussed,which lays a foundation for the wider application of the algorithm.