A practical evaluation of sequential estimation for vision metrology 1 Revised version of a paper pr

Parameter estimation in photogrammetry is generally accomplished by means of a simultaneous least-squares adjustment in which all observational data must be at hand prior to solution. It follows that, despite the rapid turnaround provided by digital imagery and current measurement technology, the simultaneous adjustment cannot offer an indication of quality until acquisition and measurement are complete. On-line quality control of single-sensor vision metrology (VM) can be implemented effectively through on-line triangulation (OLT) with sequential estimation. In combination with established network design principles, sequential estimation can have a significant positive influence on economy and productivity in the industrial environment. While enhancing the efficiency of the triangulation procedure, the risks of collecting either insufficient or surplus imagery are also diminished. Furthermore, localised weaknesses within the overall network can be isolated and corrected. Sequential estimation is applicable to a variety of VM tasks. Here, in addition to quality control for single-sensor VM, the calibration of real-time, multi-sensor systems is addressed. This paper presents an overview of sequential estimation for VM consisting of a description of the general sequential problem followed by the necessary mathematical derivations. Significant implementational aspects are discussed and evaluations of practical testing are given. Promising experimental results clearly demonstrate that OLT can be an effective and valuable tool in industrial VM.