Particle Filter-Based Robust Visual Servoing for UCF-MANUS-An Intelligent Assistive Robotic Manipulator

A particle filter based tracking scheme is proposed to robustify visual servoing of objects in the UCF-MANUS camera-in-hand vision setup. Instead of simply fusing global and local information, a concatenation of the two sources of information is proposed here which enables the combination of the two independent measurements with a synergistic collaboration between them. A novel overlap metric to encode the degree and quality of overlap between two arbitrarily shaped Regions of Interest (ROIs) is defined to facilitate the prior and posterior pdfs in the particle filter setup.A sub-ROI is defined and utilized in the observation step to facilitate the global target detection. Based on extensive experimental results under a variety of scenarios obtained by using the UCF-MANUS assistive robotic testbed, it is seen that the proposed particle filter based fusion approach is superior to other non-fused global detection or local tracking approaches. The efficacy of the proposed approach has also been verified using standard data sets. Finally, robustification of a hybrid visual servoing technique is shown by implementing the proposed particle-filter based tracker during closed-loop operation in real-time.