Monocular Vision-Based Autonomous Manipulator Insertion System

With the increasing demands for advanced automation in the fields of large-scale manufacturing industry and service industry,robots that are able to deal with complex and dexterous manipulation tasks are more demanded.The traditional hard automation is able to deal with manipulation tasks with fixed positions,fixed targets and fixed flows,however it can not deal with manipulation tasks with uncertainties,unknowns,and exceptions.To fix that problem,researchers usually install additional sensors on the manipulators to make them autonomous manipulators.Centered on the high precision insertion problems,a vision-base autonomous manipulator insertion system is proposed.A monocular camera is installed at the end of the manipulator,forming an Eye-in-Hand System.The error generated by positioning of automated guided vehicle is fixed by visual locating and visual control.This paper mainly solve two problems.Firstly,a widely applicable,markerless and monocular visual control based autonomous manipulator insertion method is proposed.The first step is using intensity-based descriptors for object detection on image plane.The second step is detecting the points on image plane,which are corresponding to the points at three-dimensional world space,and solving the perspective-npoints problem to calculate the position and pose of camera and manipulator relative to object plane.The third step is applying image registration for high insertion quality and safety before implementing insertion.Secondly,a point locating optimizing index based corner detection method is proposed to deal with the uncertain input points of the existing 3D pose estimation method.The index can be used to calculate the optimizing image preprocessing threshold.The corner detection method takes the intersections of the lines that are fitted by edges as candidate corners,then calculates the indexes of the candidate corners to find the most probable corner.Finally,we carried out the practical experiments which validated the feasibility and accuracy of the autonomous manipulator insertion method and the point locating optimizing index based corner detection method.Furthermore,we also compared the point locating optimizing index based corner detection method with classic corner detection method.