Laser radar-based range imaging with applications in shape measurement and machine control: Dissertation

The industrial applications of techniques for sensing the geometry of objects in a scene include inspection of the 3-D shapes of manufactured objects and perception of the dynamic working environments of autonomous machines. This thesis examines the feasibility of acquiring and interpreting range images produced by a laser radar to determine 3-D scene contents for such inspection and perception purposes. Various geometric sensing techniques are reviewed and the use of a laser radar technique for sensing large objects is justified. A method and environment for experimental characterization of the performance of range imaging sensors is proposed and the use of the method is demonstrated. The characterization is based on known reference objects. Two approaches to the automatic inspection of shapes of objects, one based on a raster range image and the other on profiles, are developed. The conceptual difference between the approaches lies in the way the desired data are extracted from the scene, leading to differences in sensor design and data analysis methods. The structure and operation of a laser radar-based range imaging sensor and its control and data analysis functions are defined for both approaches. The approaches are implemented and their feasibility is evaluated. The performance of the approaches and their applicability to real industrial inspection tasks are evaluated and the profile-based approach is considered as a potential solution to industrial shape inspection problems. Two perception approaches, based on either dense or sparse range imaging with a laser radar, are discussed, the amount of data collected being held to distinguish the approaches and to lead to significantly dissimilar implementations and characteristics. No final solution is proposed here, but their feasibility is evaluated based on experiences gained with subsystems implemented and demonstrated in this work. From the utilization point of view, the two approaches seem to be complementary rather than competing and possess different degrees of maturity. Guidelines are suggested for further research needed to confirm the present conclusions.