Adaptive edge detection and line extraction for the Vision Survey System

Brouwer, A.R.; Adaptive edge detection and line extraction for the vision survey system. M.Sc. Thesis, Measurement and Control Section ER, Department of Electrical Engineering, Eindhoven University of Technology, The Netherlands, June 1994. A Vision Survey System is being designed as a part of an intelligent robot welding system. The VSS uses CCD cameras to generate a geometrical scene description with the aid of knowledge based object model descriptions. Subtasks of the VSS are camera calibration, low-level vision operations, structural and geometrical matching and object part clustering. A synthetic image modeler is applied for verification purposes. The subjects of this thesis are the low-level vision module algorithms and the design and implementation of a user interface for the VSS. Output of the low-level vision module is a set of line-segments. These are obtained from a grey-level input image by performing lens correction, edge detection, line extraction and post-processing in succession. The task of edge detection is marking locations where changes in the brightness of the image occur. The line extraction algorithm attempts to fit straight lines through these so-called edgepixels. Schemes are proposed to automatically set optimal values for some edge detection parameters. Application of a post-processing algorithm condenses the number of lines and enhances their accuracy. The edge detection algorithm was improved by combining brightness information from different images, obtained with different lighting conditions. A user interface has been implemented using MS-Windows 3.1. A separation has been made between the coding of the VSS algorithms (nonwindows) and their graphical output and parameter settings. These I/O functions are handled by the user-interface.