Digitizing automotive production lines without interrupting assembly operations through an automatic voxel-based removal of moving objects

We present an efficient method to partition a point cloud gathered through kinematic laser scanning into static and dynamic points. The presented algorithm utilizes a voxel grid data structure and uses a ray intersection test to mark voxels as dynamic. The algorithm does not require any ego-motion estimations, computationally expensive object recognition or tracking of moving objects over time. It is easy to implement and can be executed on many cores in parallel. We show the viability of this approach by applying our algorithm to a dataset that we gathered by mounting a FARO Focus3D Laser scanner onto a skid which was then sent along a production line for consumer car chassis in a factory of the Volkswagen corporation. Since factory operators are interested in acquiring digital models of their production lines without suspending factory operations, the resulting point cloud will contain many dynamic objects like humans or other car bodies. We show how our algorithm is able to successfully remove these dynamic objects from the resulting point cloud with minimal errors. Our implementation is published under a free license as part of 3DTK.