SLAM with Laser Profilers for High Definition Mapping in Confined Spaces

Abstract

Three-dimensional reconstruction in confined spaces is important for the manufacturing of aircraft wings, the inspection of narrow pipes, the examination of turbine blades, etc. It is also challenging because confined spaces tend to lack a positioning infrastructure. Therefore, a sensor that is capable of performing Simultaneous Localization and Mapping (SLAM) is required. Although there exist a variety of SLAM-capable sensors such as LiDARs and RGB-D sensors, there have been few, if any, sensors for confined spaces reconstruction, because such tasks require sensors that are compact, operate in short-range, and can self-localize.

In this thesis, we propose a sensor framework based on monocular laser profiling for confined spaces. This framework consists of a hardware structure, a software pipeline, and a SLAM method. Sensor prototypes designed using this framework are able to achieve photo-realistic 3D reconstruction in real-time despite a monocular sensor setup. A SLAM method tailored to laser profilers is proposed to accurately localize the sensor by tightly fusing laser, camera, and inertial measurements. This sensor framework's ability to generalize to different sensor configurations enables it to tackle various confined spaces. For the general confined space setting, the Blaser prototype features a laser-stripe profiler and was designed to be compact and short-range-capable. It boasts a 1-inch minimum sensing range and is more than ten times smaller than Intel RealSense D435. For confined in-pipe environments, a more specialized prototype named PipeBlaser is designed. It has a laser-ring profiler configuration and can function in 12-inch diameter pipes. These two sensor prototypes exhibit vastly different configurations but are designed under the same sensor framework with some modifications for corresponding applications.

A comprehensive qualitative and quantitative evaluation was performed on both sensor prototypes in a variety of environments, demonstrating their localization and mapping capability in a real-time fashion. We also compare the sensor system to other state-of-the-art SLAM methods as well as to a popular and capable RGB-D camera.