This paper presents a novel method for free-form registration of multiple point clouds. The method adopts a parallel-light data association design inspired from torchlight structure which improves the correctness of point correspondence. When two sets of point clouds are placed together, assume a set of parallel light beams are passing through them. Each light beam will pass the point clouds twice, one on each data set. The Euclidean distance on each light beam between the two sets are taken as measurement of the separation. The fitness is the reciprocal of the mean distance of all light beams. When the two sets are optimally aligned, the fitness is maximized. Hence, the registration problem is reduced to a six degree of freedom search. Preprocessing and acceleration modules such as Genetic Algorithm (GA) are introduced to reduce the exploration space and execution time. Unlike the Iterative Closest Point (ICP) algorithm, the proposed algorithm does not require pre-alignment information. Secondly, ICP does not perform well when the overlapped area between two sets are not sufficiently large. And the proposed algorithm does not suffer from this partial overlapping problem. Based on various experiments with real data, the proposed method has superior performance compared to ICP.
Swipe to navigate through the chapters of this book
Please log in to get access to this content
- A Novel 3D Registration Algorithm Using Parallel-Light Association
- Copyright Year
- Springer Berlin Heidelberg