Mosaicking images: Panoramic imaging for miniature robots

Two methods of mosaicking images were developed, one for the general case and one for the simple case of a shift between two following images. The general method was evaluated by merging three images into a mosaic. When executed manually, the success rate of the method was 60%. With human assistance in selecting an appropriate number of corners to base the correlation on, the success rate increased to 89%. The operation took four minutes on a Pentium II 450 MHz Linux workstation, with most of this time spent on finding the correspondences. For the simple method, the shift was determined correctly in 95% of the tests runs for less noisy images. For noisy images, the success rate of properly merging all images of a mosaic was 20%. To increase the success rate, a function that determines the noise level was added. If too much noise was detected, the image is discarded and another one taken. This way, the error rate dropped to 10%.