Autonomous indoor mobile mapping has opened up new horizon in the field of surveying and mapping industry. The ability to create a 3D map without user intervention not only reduces labour costs but also provides more flexibility for exploring remote sites. Hence, it is worthwhile to consider the role of robotics in the mapping industry.
The primary demand for autonomous robot systems is to interact with environment for obstacle avoidance and self-localization in six degrees of freedom (x-, y-, z-position, roll, yaw and pitch angle). The later issue requires knowledge of the operating environment, which leads to automatic environment modeling or environment mapping solution.
Two different scenarios for autonomous indoor mobile mapping are investigated in this thesis. The first scenario is based on the use of a single RGB-D sensor to map a small room of size (8x8 meter). In the second scenario RGB-D sensor is used as an aiding sensor for Velodyne HDL-32 LiDAR to map a large corridor of size (33x11 meter). The results shows that the solution of single RGB-D sensor is accurate enough for mapping a small room; however, for large corridor the result of RGB-D aided Velodyne HDL-32 generated more accurate and consistent mapping solution.
The main challenge that should be handled for autonomous mapping is alignment of multiple local scans as they become locally distorted because of the motion of the platform and noise in sensor measurements. The collected scans from multiple locations are associated with the individual sensor locations (the capturing process is done using stop-and-go approach, where the robot is stopped at different locations to capture the scene). Hence, a registration process must be performed in order to combine several scans at different locations. The main goal of the registration process is to estimate the transformation parameters, which will define the relation between the collected datasets from different locations.
The optimization and enhancement of the registration procedure plays a major role for generating indoor mobile mapping solution. The problem of alignment is addressed through several optimization steps, starting from coarse registration, followed by fine registration, segmentation and finally loops closure.