Dual supervisory architecture for drift correction and accurate visual servoing in industrial manufacturing

Pose and motion estimation techniques required to drive robotic manipulators involved in quality control in the automotive industry often encounter important limitations when applied on weakly textured panels moving in complex environments. Difficulties originate from the reduced performance of classical feature extraction, tracking and matching algorithms that heavily depend on the presence of rich textures over the objects. While maintaining a simple hardware architecture relying on passive stereoscopy, this paper proposes a supervisory approach for machine vision systems that overcomes the sensitivity of feature-based tracking to occlusions, photometric variations and ubiquitous appearances of factory associates in the field of view of the vision system. Furthermore, the drift accumulating with the tracking module is also corrected. Experimentation demonstrates the suitability of the proposed supervised pose and motion estimator for automated robotic marking of surface deformations over moving panels.