Novel IBVS System Design for Cable-Driven Hyper-Redundant Manipulator With MLESAC-Based Feature Vector Optimization

The visual servoing control extends the potential application of cable-driven hyper-redundant manipulators (CDHRM) to automatically execute tasks by sensing the visual information of environment. However, the high proportion of feature point mismatching and visual occlusion may occur in the complex and narrow environment that CDHRM works in, which make the traditional image-based visual servoing (IBVS) system hard to work or even cause instability. In this article, a novel IBVS system is proposed for CDHRM with eye-in-hand configuration, where the ability of mismatching resistance is improved by optimizing the vector of feature points, while the stability and tracking performance are still guaranteed. The maximum likelihood estimation random sample consensus (MLESAC) algorithm is designed to estimate the inlier model that resists mismatching and extracts well-matched feature points (namely the inliers) for the first captured image. Since the inliers may become mismatched in the newly captured image, the inliers and inlier model are both updated according to the feature quality weights. As a result, the newly mismatched feature points are excluded so that the weighted feature vector with less mismatching and higher quality can be generated. Subsequently, the weighted IBVS control law based on this vector is designed to achieve the mismatching resistance as well as guarantee the asymptotic stability and tracking performance of system. Comparative experiments are implemented for the proposed MLESAC algorithm and novel IBVS system, and the results verify that our method has better adaptability to the environment when applied in CDHRM, even with partial visual occlusion.