| Résumé : |
(Auteur) The advent of filmless imaging systems, especially of Charge-Coupled Devices (CCD), has created manifold opportunities and new applications which have led to significant changes in Photogrammetry. The exacting demands of Photogrammetry on the radiometric and geometric characteristics of the imaging sensor and all other elements involved in the acquisition of imagery with solid-state sensors, require a detailed analysis of the factors affecting the performance.
This analysis can build on extensive knowledge acquired over the last twenty years for the calibration of film based cameras. For example the modelling of systematic errors introduced by lenses is identical for both systems. Calibration and analysis techniques, such as the bundle adjustment with self-calibration, are important tools. Another source of a great deal of information are applications of solid-state sensors in astronomy. Exacting radiometric requirements have led to the development of specialized sensors, cameras, image acquisition systems, and calibration techniques. These techniques are currently only partially applicable to photogrammetric tasks though. They are tuned to be used with extremely specialized hardware (cameras, image acquisition systems, testing arrangements) which is typically not available for photogrammetric applications. One task of this dissertation was thus to investigate the effects of cameras, signal transmission techniques, and frame grabbers as they are widely available on the market as off-theshelf equipment. The individual components were investigated first, then the combined radiometric and geometric performance of the system was analyzed, and finally the performance of a system for three-dimensional measurements was verified. This in turn required the installation of an image acquisition system with advanced capabilities to perform various comparative investigations, to set up a three-dimensional testfield, and to write a large amount of software for analysis and calibration.
The geometric regularity and the excellent radiometric characteristics of solid-state sensors make them ideal measurement devices. A number of investigations have shown that the regularity of the sensor element spacing is in the order of 11100 of the spacing. The uniformity of response from sensor element to sensor element is 1 % and better for many off-the-shelf cameras. The geometric regularity of the sensor would thus allow to measure positions of targets imaged on such a sensor with a precision of I/100th of the spacing and better. This is actually achieved and surpassed with special sensors in star tracking applications. The mechanical design and the electronics of CCD-cameras are usually not designed for photogrammetric purposes. The mechanical design is often not sufficiently stable, i.e. the assembly of lens, housing and sensor is not rigid. The camera electronics are designed for visual purposes (e.g. surveillance), thus potentially introducing significant degradations due to a number of factors such a low-pass-filters, gamma correction, addition of video signals, to name but a few.
The largest drawback with respect to the radiometric and geometric performance, but an advantage with respect to economy, is the use of standard analog video signals for the transmission of the imagery acquired by solid-state sensors. These standards were developed in the 1950's for broadcasting and are not at all tuned for the requirements of precise measurements. The radiometric and geometric properties of these signals were analyzed, potential drawbacks discussed, as well as methods for determination and compensation and/or elimination of deficiencies investigated and performed.
Frame grabbers are another critical component of the image acquisition system. They must convert the analog transmitted imagery into matrices of numbers, the digital image.
The radiometric and geometric performance of a number of electronic components and of different synchronization techniques of the frame grabber were analyzed. These tests revealed a number of critical issues such as proper DC-restoration, degrading effects of low-pass-filters, deviations from linearity, low level patterns, and the performance of pixel-synchronous sampling as compared to PLL line- synchronization. This allows to separate influences of the frame grabber from those of the camera. The assessment of the effects of the two synchronization techniques, pixel-synchronous sampling and PLL line-synchronization, lead to the detection of a geometric deformation (which can be introduced when composite video signals are used in connection with PLL line-synchronization) and the confirmation of the degrading effects of line-jitter. It was furthermore shown that pixel- synchronous sampling provides a transmission without loss for photogrammetric purposes. The investigation of several radiometric characteristics of the complete system demonstrated some of the insufficiencies of typical off-the-shelf hardware. Potential trouble spots, such as degradations of the radiometric uniformity at the borders of the imagery, were detected. The difficulty of radiometric calibration of the system was addressed.
A test strategy for the assessment of the geometric stability of the system was developed. The effects of the synchronization on the radiometric performance was used to develop test methods to rapidly pinpoint imprecisions of the frame grabber synchronization. The analysis of the repeatability proved that pixel- synchronous frame grabbing provides identical geometric characteristics as digital transmission. The short time repeatability for both digital image transmission as well as pixel-synchronous sampling was shown to be in the order of 0.004 pixel. The analysis of PLL line--synchronization using pixel-synchronous frame grabbing as reference confirmed that line jitter reduces the internal precision in row direction by a factor of six. Disturbances of the PLL line-synchronization, introduced by composite video signals, were shown to lead to geometric deformations reaching 0.3 pixel. The study of warm-up-effects allowed to determine the minimal time required by the system (camera and frame grabber) to reach a steady state. When using pixel-synchronous sampling the major contributing factor to warm-up effects are thermally-induced deformations of the camera body, i.e. the assembly of lens and sensor. The geometric stability of the image acquisition system was verified with a test of over one week duration. It could be shown that for durations of several days a repeatability of well below 1/100th of a pixel can be achieved. Finally the effects of local variations of the illumination andlor of shadows on the position determination with Least-Squares Matching were analyzed. It was empirically demonstrated that gradients of a few per cent can lead to errors in the estimated position of several hundred's of a pixel.
The geometric precision and accuracy of the system was verified with a three-dimensional testfield. The part of the testfield used in the tests spans a volume of 2600 x 2000 x 1100 mm with 162 targets of 20 mm diameter. The position of the targets was determined with a precision of 0.02 to 0.03 mm by theodolite measurements. Two sets of 48 frames were acquired, one with pixel-synchronous sampling and the other with PLL line-synchronization. The relative accuracy attained with the pixel- synchronously grabbed imagery reaches 1 part in 50 000 in object space and is 1/85th of the pixel spacing in the image when using 30 control points were used. This is reduced to 1 part in 46 000 and 1/50th of the pixel spacing with a minimally constrained network. The accuracy attained with the imagery acquired with PLL line- synchronization is only slightly lower. It was thus shown that very high accuracies can be attained with PLL line-synchronization when appropriate modelling of the geometric deformation and a large number of images are used. A comparison of the object coordinates computed with the two data sets indicated a relative accuracy of 1 part in 70 000 and 60 000 for the X and Y axes respectively. The accuracy in image space of the comparison is 1/100th of the pixel spacing. Finally some factors which limit the accuracy and approaches for their elimination are discussed. |
Geometric and radiometric analysis of a CCD-camera based photogrammetric close-range system (1992)
