Backpack-based inertial navigation and LiDAR mapping in forest environments

Abstract: Creating 3D models of our surrounding world has seen a rapid increase in research and development over the last few years. A common method is to use laser scanners. Mapping is done either by ground based systems or airborne systems. With stationary ground-based laser scanning, or terrestrial laser scanning (TLS), it is possible to obtain high accuracy point clouds. But stationary TLS can often be a cumbersome and time-demanding task due to its lack of mobility. Because of this, much research has gone into mobilised TLS systems, referred commonly to as mobile laser scanning (MLS). Georeferencing point clouds to a world coordinate system is a difficult task in environments where global navigation satellite systems (GNSS) is unreliable. One such environment is forests, where the GNSS signal can be blocked, absorbed or reflected from the trees and canopy. Accurate georeference of points clouds for MLS systems in forests is difficult task that can be solved by using additional measurement instruments and post-processing algorithms to reduce the accumulation of errors, also known as drift. In this thesis a backpack-based MLS system to be used in forests was tested. The MLS system was composed of a GNSS, an inertial navigation unit (INS) and a laser scanner. The collected data was post-processed and analyzed to reduce the effects of detecting multiple ground layers and multiples of the same tree due to drift. The post-processing algorithm calculated tree and ground features to be used for adjusting the point cloud in the horizontal and vertical planes. The forest survey was done for an area roughly 40 meters in diameter. The MLS data was compared against TLS data as well as manual caliper data - where the caliper data was only measured in an area roughly 24 meters in diameter. The results indicated that the effects of multiple ground layers and multiple tree copies were removed after post-processing. Out of the total 214 TLS trees, 185 managed to be co-registered to MLS trees. The root mean square error (RMSE) and bias of the diameter at breast height (DBH) between the MLS andTLS data were 27.00 mm and -9.33 mm respectively. Co-registrationof the MLS and manual caliper data set gave 36 successful matches out of the total 43 manually measured DBH. The DBH RMSE and bias were 16.95 mm and -10.58 mm respectively. A Swedish TLS forest study obtained a DBH RMSE and bias (between TLS and caliper) of approximately 10 mm and +0.06 mm respectively. A Finnish backpack MLS forest study obtained a DBH RMSE and bias (between MLS and TLS) of 50.6 mm and +11.1 mm respectively. Evaluating the difference in radius at different heights along the tree stems between the MLS and TLS revealed a slight dependence on height, as the radius difference increased slightly closer to the stem base. The results indicated that backpack-based MLS systems has the potential for accurate lidar mapping in forests, and future development is of great interest to improve this system further.