UAV-Based Multi-Sensor Integration for Precision Navigation and Object Classification

In this dissertation, a UAV-based multi-sensor data fusion for precision navigation and object classification for seamless indoor-outdoor applications was developed. A UAS equipped with a GNSS/IMU board, a low-cost mechanical LiDAR sensor, and an RGB camera, was used to collect data for the purpose of developing a multi-sensor integration system. Unfortunately, due to a malfunctioning GNSS antenna, there were numerous prolonged GNSS signal outages. To resolve this issue, a GNSS/INS/LiDAR-integrated system was developed. A LiDAR SLAM solution was used to update the INS solution through an EKF. Two case studies were considered: a complete GNSS outage and a GNSS PPP-assisted solution. In comparison with the complete GNSS outage, the results for the second case study were significantly improved. Subsequently, a low-cost UAV-based multi-sensor data fusion model was developed for land cover classification using a DCNN. The collected dataset was used to train and test the developed DCNN. In order to validate the DCNN, a second dataset was collected at a different urban location using a solid-state Lidar UAV-based system. It was shown that the proposed DCNN approach improved the overall accuracy of land cover classification significantly compared to the reference classifiers for both datasets. The feasibility of fusing UAV-based imaging and low-cost LiDAR data was explored to enhance pavement crack segmentation using a DCNN model. A third dataset was collected using the same UAS used in collecting the second dataset. Two types of pavement distress were investigated using the first and third datasets, namely cracks and sealed cracks. The performance metrics decreased for the crack samples in both datasets when the LiDAR data was added. This was essentially due to the quality of the lower-grade LiDAR sensor, which had low spatial resolution. In contrast, for the sealed crack, the addition of LiDAR data improved the matrices for the segmentation accuracy. Finally, an event-based visual-inertial odometry approach was proposed for precise indoor localization. The proposed approach fused events, standard frames, and inertial measurements. The proposed approach resulted in a significant decrease in the RMSE compared to the event-only and the frame-only cases, respectively.