CMSC 2024

Photogrammetry using off the shelf low-cost camera systems is being applied to a wide variety of industrial activities where the emphasis is on determining the position, pose and potentially deformation of multiple objects within the field of view of the combined cameras over increasingly large volumes at sub-mm levels of uncertainty. Understanding and calibrating the internal imaging geometry of these camera systems to correct for systematic errors whilst maintaining the freedom to deploy different sensors and lenses with varying angles of view and image magnifications is a key enabler to meeting industrial requirements. This presentation will focus on getting the most from established bundle-adjustment based camera calibration methods. Such methods form a routine part of our low-cost multiple camera photogrammetric system deployment on tasks concerned with tracking robots, tools and objects under both laboratory and industrial situations.

Off the shelf cameras, are typified by CMOS imaging sensors purchased either as printed circuit board level devices with small threaded lens mounts, as physically enclosed units with “C” mount and larger lens mounts packaged for working in harsh environments, or as stand-alone consumer photographic cameras operating under their own power with local camera control and storage capability. No matter what type of imaging system is selected, it is necessary to numerically model the systematic geometry of the photogrammetric light path from object space to image space. The widely accepted approach of using a bundle adjustment solution permits the combination of measurements made from one or more camera images and their uncertainties along with a reference frame definition and photogrammetric lens model to estimate coordinates, positions, and poses of objects which are in the combined viewing volume of the cameras when the image set was taken. The bundle adjustment process provides a rigorous least squares error propagation enabling the uncertainties of the required location, pose to be estimated.

This presentation will describe rigid object and scene-based camera calibration methods which can be deployed in either laboratory or factory spaces illustrating results with examples drawn from our research. We will include digital consumer cameras imaging with different generations of moderate wide angle lenses typical of “off-line” or single mobile camera photogrammetry highlighting the importance of 3D spaces and calibration objects, matching camera image quality with the features being used to support the calibration process, suitable image network geometries and image coverage to achieve a reliable calibration, the parameters from bundle adjustment that give most insight into the success of the calibration process. We will also look at the calibration results from a range of mid-range quality lenses optimised for imaging with the 20MP generation of “C” mount camera units considering camera calibration outputs with respect to different lens angles of view under constant magnification. The outcome will be a set of best practice hints, tips and tricks necessary to achieve reliable and accurate photogrammetric camera calibrations. The tables below give examples of some of the areas that will be discussed: (1) variation in target image quality with lens aperture and location in the image; (2) estimated calibration parameters for five different lenses fitted successively to the same camera body; (3) variations in the internal corelations between estimated camera calibration parameters for a near zero radial distortion lens; (4) image residuals following bundle adjustment overlaid from all images taken with a given camera and (5) 3D estimated coordinate uncertainty ellipsoids estimated from bundle adjustment output covariance matrices.