Projects

dbicp

DBICP projectAbstractThe DBICP Project studies a point based registration algorithm called the “Dual Bootstrap Iterative Closest Point”. This algorithm extends the classic Iterative Closest Point (ICP) algorithm, to overcome issues such as initialization sensitivity, few overlap, and ... [More] unreliable matches. The innovations are made in the algorithm's structure, where the region used – the bootstrap region –, and the parametric transformation model selected are progressively “bootstrapped”, meaning enlarged. Finally, my source code is available (for free!) on this website! Full reportAvailable here! Full presentationAvailable here! DemosBasic ICPHere are a two examples, illustrating a stuck-in-local-min and successful optimizations using the Basic Iterative Closest Point algorithm. This shows the sensitivity to initialization. Model BoostrapHere are a two examples: in the first video, the parametric transformation is forced to be quadratic (no model bootstrap). in the second video, the algorithm automatically selects the transformation model (model bootstrap activated). The results is clearly better, even if not perfect, but it illustrates well the idea. [Less]