Saliency Based Fast Object Localization and Recognition for Mechanical Assembly

This paper presents a fast object localization and recognition approach for mechanical assembly based on detection of multiple salient regions. Our novel multiple salient object detection technique employs locally adaptive center-surround operations and proto-object partitioning. The proposed approach is implemented in a vision guided industrial robot workspace to perform assembly involving different types of components. Effectiveness and utility of the proposed approach in robotic assembly is demonstrated through different measures obtained from multiple experiments. Detection and recognition rates, angular and positional errors, and computation time of our implementation are found to be better than when multiple salient object detection is not considered.