Object Recognition and Localization from Scanning Beam Sensors

Model-based object recognition and object localization are fundamental tasks in industrial automation. In this article, we present a system that quickly recognizes (5 μs) and accurately localizes (0.025 mm) objects using a scanning beam sensor that consists of an array of binary light-beam sensors. Our scanning beam sensor is robust, inexpensive, compact, precise, and insensitive to ambient light, which are all prerequisites for industrial manufacturing applications. Scanning beam sensing involves moving objects with respect to the sensor and record ing the manipulator positions when the sensor outputs change. The recognition problem and the localization problem share the correspondence subproblem, the task of interpreting the sensed data in terms of model features. In this article, we present a constant-time indexing correspondence algorithm. Indexing in volves discretizing the sensed data to achieve integral indices, and using these indices to look up a table entry containing the correspondence information (the consistent model feature inter pretations). Complete indexing tables are crucial for scanning beam sensing, and for sparse sensing strategies in general, because each experiment only produces a handful of indexing coordinates. Constructing complete indexing tables has previ ously been an open problem (Clemens and Jacobs 1991). In this article, we describe a method for constructing complete indexing tables by enumerating the cells in an arrangement in configuration space.