The concatenation of transformations in the context of video mosaicking or 3D mapping of environments features an inevitable drift due to the accumulation of random and remaining systematic errors. These drifts become obvious in the form of gaps appearing in areas of overlap which are obtained by the repeated acquisition of a scene. An efficient method for the adjustment of multiple loops established by place recognition is introduced. The equal distribution of the gaps among the transformations yields consistent mosaics and drift reduced sensor trajectories respectively in 3D. Potential applications are video mosaicking, image stabilization or consistent mapping of 3D environments.
Video mosaicking by adjustment of multiple loops
Correcting of a drifting sensor motion. Left: uncorrected sensor trajectory. Right: adjusted motion (3 loops)