Least Squares Matching Method: Some Experimental Results

The effect on accuracy resulting from some changes to the conventional least squares matching method is investigated. Parallaxes computed automatically with matching methods have been compared with manual measurements made in a stereocomparator. Examples have been taken from both large and small scale aerial photographs and from close range images. Different linear geometric and radiometric parameters have been used as additional unknowns. The use of data snooping in digital matching with the least squares matching method has also been investigated. The results show that the introduction of additional affine parameters has a positive effect on the accuracy when the window size is larger than 30 × 30 pixels. The use of the data snooping technique is promising. The best precision obtained was between 10 pm and 15 μm, expressed as radial root mean square deviation, corresponding to approximately half a pixel. Gross error detection is possible to a large extent with photographs of good image quality.