A Practical Study of Gross-Error Detection in Bundle Adjustment

A special bundle adjustment program has been developed with a built-in automatic gross-error detector which uses the data snooping approach. The program computes, in an efficient way, the exact value of the redundancy numbers for each image point. Using actual data, many of the factors affecting the reliability of bundle adjustment and the ability of the technique to detect gross errors are studied. The effects of different block parameters, point type and location, and the use of additional constraints are presented.