Reliable Real-time Change Detection and Mapping for 3D LiDARs

Abstract

A common scenario in Search and Rescue robotics
\nis to map and patrol a disaster site to assess the situation and
\nplan potential missions of rescue teams. Particular importance
\nhas to be given to changes in the environment as these may
\ncorrespond to critical events like building collapses, movement
\nof objects, etc. This paper presents a change detection pipeline
\nfor LiDAR-equipped robots to assist humans in detecting those
\nchanges. The local 3D point cloud data is compared to an
\noctree-based occupancy map representation of the environment
\nby computing the Mahalanobis distance to the closest voxel in
\nthe map. The thresholded distance is processed by a clustering
\nalgorithm to obtain a set of change candidates. Finally, outliers
\nin these sets are filtered using a random forest classifier.
\nChanges are continuously mapped during a sortie based on
\ntheir classification score and number of occurrences. Changes
\nare reported in real time during robot operation.