Stereoscopic line-scan imaging using rotational motion
Richard S. Petty · Nottingham Trent University's Institutional Repository (Nottingham Trent Repository) · 1997
This thesis describes work which has been carried out to investigate a rotating stereoscopic imaging system consisting of two line-scan cameras. Algorithms have been proposed and experimentally tested which allow three-dimensional co-ordinate information to be derived from a defined object space. A rotating stereoscopic line-scan system would be particularly suited to applications in which information is required from the "all-round" observation of a scene. Such applications would include autonomous vehicle guidance, path-planning for complex manipulator manoeuvres or security surveillance scenarios. The ability to extract three-dimensional co-ordinate information from the stereoscopic field of view would allow a solid image model of the workspace to be constructed. Initial work involved a theoretical appraisal of a line-scan camera used in a rotating two-dimensional mode. Such a system was constructed and experiments were undertaken to determine the suitability of the rotating sensor as the basis for the development of a rotating stereoscopic camera arrangement. The results from this section of the work verified that two-dimensional co-ordinate information can be obtained from a defined object space. Subsequently, a rotating stereoscopic line-scan system was constructed for which a theoretical mathematical model for measurement was developed. Application of the derived algorithms to the stereoscopic system enabled the position of targets in three-dimensional object space to be determined on completion of a calibration procedure. Experimental work was conducted to quantify any measurement errors inherent in the stereoscopic configuration and to identify, if possible, the source of these errors. Further experiments were undertaken to determine the accuracy of measurement which could be achieved using the algorithms developed. The results indicate that the three-dimensional position of targets in object space can be determined, at best, to an accuracy of ±0.5mm in the X-axis (horizontal), ±0.6mm in the Y-axis (vertical) and ± 1.2mm in the Z-axis (range) at a camera-to-object range of 1.5m.