Industrial part recognition using a component-index

摘要

Part recognition is an important problem in industrial vision. Efficient and accurate part identification is essential for flexible automation of almost all major manufacturing processes such as inspection, assembly sorting and binning. The problem of 2D industrial part recognition is considered — a new data-driven technique for part recognition is proposed. This technique models an object or a scene as a composition of several components. A component corresponds to a convex section of the object boundary. A component of the unanalyzed portion of the scene is efficiently identified using a k-d tree-based component-index. The objects that contain the identified component are hypothesized to be present in the scene. A given hypothesis is verified by matching the transformed boundary of the hypothesized object against the scene. Some experimental results are also discussed to demonstrate the efficacy of the proposed technique.