Area-based least squares matching has existed since the 1980s, and its development may therefore be regarded by some photogrammetrists as complete. However, this is a relatively short time in a rapidly changing discipline, and the method does have scope for further refinement. This paper aims to enhance the fidelity of the fundamental mathematical model, which relates pixel positions in area-based matching. If successful, its benefit is seen to lie within close-range photogrammetry, in which object undulations, convergent camera directions and large image scales can introduce significant levels of perspective distortion. It is hypothesised that a model, which takes into account the surface shape within the match window, would provide a pixel position relationship which is applicable across larger windows than those which are applicable with the conventional matching model, based on an affine transformation. The use of larger windows for the image matching increases the redundancy. Revised co-ordinate transformations, based on mathematical surface models across the windows, are proposed in this paper. When tested on the measurement of real objects, even simple surface models are found to increase the complexity of the matching mathematics, but when compared with the traditional affine transformation solution using three test objects, precision improved noticeably. Generally, accuracy also improved, but the improvements were not as distinct as they were for precision. Quicker convergence with fewer iterations was usually obtained, and this is seen as particularly indicative of a more faithful model. Yet more rigorous surface modelling may be worth developing, but the means of choosing the most appropriate models for different objects also remains a question deserving to be pursued.
ISPRS Journal of Photogrammetry & Remote Sensing Vol. 56, Issue 1, p. 42-52