Automatic Configuration of an Object Detection System

Abstract:
A novel configuration method for pre-processing steps required to segment range image data previous to object recognition by geometric primitives is presented in this paper. Pre-processing in this context denotes steps of smoothing, segmenting and sorting the image into meaningful regions. The resulting regions serve as input for the following detection of the pose of freely tangible parts. The generation and configuration of the pre-processing steps, which are subsumed as pre-processing strategy, is done using an evolutionary optimization method using 3D results to evaluate the quality of the results of the 2D image processing based pre-processing strategy. The pre-processing is build up fully automatic without user interaction and leads to real-world applicable solutions.