This thesis presents a comprehensive framework for an obstacle-avoiding autonomous unmanned aerial vehicle (UAV) system with a focus on structural health monitoring (SHM) in global positioning system (GPS)-denied areas. The proposed framework integrates a new obstacle avoidance method (OAM), a localization method using fiducial ArUco markers, and a real-time crack segmentation method. The OAM utilizes You Only Look Once version 3 (YOLOv3) network and a K-means clustering algorithm for robust obstacle detection and clustering. The ArUco marker-based localization method overcomes the limitations of traditional ultrasonic beacon (USB) localization, providing reliable and accurate UAV localization even in the presence of magnetic interference. Comparative studies show that the ArUco marker-based localization method significantly reduces yaw control error by 60.45% and path following error by 67.29% compared to USB-based localization. The developed autonomous UAV system is implemented and validated in both indoor and outdoor environments, demonstrating its effectiveness in GPS-denied areas. Furthermore, the integration of a state-of-the-art crack segmentation network (STRNet) enhances the system's capability for real-time crack segmentation with superior performance (mIoU 92.5%) compared to other deep convolutional neural networks.