The Analysis of Engineering Drawings using Robot Mapping Techniques

Engineering drawings provide a significant challenge to image analysis. The goal is to take a scanned engineering drawing image and produce an interpretation of the contents in terms of characters, digits, arrows, line segments, dimensions, etc. These can then be incorporated into a legacy system re-engineering process (e.g., image analysis must provide parameter values for manufacturing features like counterbore holes, etc.). We propose to treat the problem in two steps: (1) determine a good set of feasible points which constitute the markings on the page, and (2) subsequently treat these as a 2D floor plan that is explored by tiny mobile agents that navigate through the drawing and produce a map. The trajectories followed by the agents allow a segmentation of the image into basic geometric units: line segments, end points, branch points, etc. This method is attractive because a large set of robust techniques have been developed. Moreover, learning techniques may be employed to determine characteristic features of the line drawing data. The thesis of this work is that mobile robot mapping techniques can be used to achieve accurate and robust segmentations of engineering drawing images.