Robot Visual Servoing Control Based on Redundant Features of Sub-pixel Accuracy

To overcome the dependency of a robot visual servoing system on calibration accuracy and pose estimation, a redundant featurebased robot visual servoing control method was proposed via fusing object feature recognition and visual servoing in different stages respectively. Firstly, to address the issue of largescale computation on image processing, a recursive greedy data compression algorithm based on curve vector was designed to speed up the feature extraction process. Secondly, to improve the measurement precision of feature in image space, a subpixel feature extraction method was studied based on the principle of vector orthogonallity. Furthermore, experimental rule of object recognition was proposed based on cooperative object shape and polygon shape fitting. Finally, based on the theory of image visual servoing and the method of task function, redundant visual features in subpixels were selected directly as feedback signals to build controlling model of robot visual servoing. The theoretical analysis and experimental results show that the servoing features can be extracted quickly and stably in complex environments, and the proposed method was robust to the calibration error and depth estimation error.