Algorithmic Approach to Planar Void Detection and Validation in Point Clouds

When using exploratory robotics for nuclear decommissioning, the generation of accurate and complete maps of the environment is critical. One format these maps can take is point clouds. However, when generating point clouds it is likely that voids, resulting from shadows created by features within the environment for example, will exist within them. Previous research studies have developed techniques which enable such voids to be interpolated. Unfortunately, for hazardous environments, this interpolation can be detrimental and have severe safety implications. This paper proposes a new algorithmic method for simplifying the detection of voids in point clouds. Once detected, the voids are validated, to make sure they exist within the scan-able part of the environment and if they do then they are marked for further investigation by the robot. This enables a more complete point cloud to be generated. To demonstrate the capabilities of the method, it was initially applied to a set of simplified scenes, where it was able to detect all the voids that were present, whilst also being compared to another algorithm. Following this the method was applied to a more realistic scenario. In this example, many, but not all of the voids were correctly identified. A discussion follows explaining why some voids were not detected and how future research aims to address this.