Advanced point cloud generation for photogrammetric modeling of complex 3D objects

A novel method using self-adaptive image correlation windows for photogrammetric point cloud generation is reported. By combining existing tie points used in the bundle adjustment with a flood-fill matching process that uses a tie point as a seed, the algorithm makes it possible to develop point clouds and surfaces automatically for a variety of complex real world scenes. The performance assessment is on-going, but first results show that, although computationally expensive, the method can be utilized in a practical setting with little operator intervention. Future work will focus on improving the adaptive windows and on fitting various surfaces during matching so that composite object types can be sent to the CAD environment as an alternative to point clouds, which are expensive to manipulate.