EVALUATION OF THE QUALITY OF REAL-TIME MAPPING WITH CRANE CAMERAS AND VISUAL SLAM ALGORITHMS

Abstract

In the context of the development of an autonomous tower crane, the usage of crane cameras to map the respective workspace as basis for autonomous path planning is investigated. The goal is to generate an up-to-date DEM as input for the crane control. As construction sites are highly dynamic scenes, it is crucial to be able to react to any changes. Thus, real-time mapping with a visual SLAM solution is aspired. As the quality of the DEM is important for such a safety critical application, we are evaluating the mapping quality of four state-of-the-art SLAM solutions, namely ORB-SLAM3, LDSO, DSM and DROID-SLAM. The results show that all approaches can handle our specific crane camera setup and thus are generally suited for our application. The DEM accuracies of all tested methods are even competitive with the result from standard offline photogrammetric processing, at least for the major part of the test site. However, there are limitations with regards to the DEM completeness. Consequently, our investigations show that the tested methods deliver a good basis for real-time accurate mapping, but for their application for autonomous path planning further refinements have to be made.