Date of Award

Summer 8-27-2021

Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering and Computer Science


Velipasalar, Senem


aerial vehicle, autonomous, computer vision, point cloud, thermal anomaly detection, UAV

Subject Categories

Artificial Intelligence and Robotics | Computer Sciences | Physical Sciences and Mathematics


Undoubtedly, unmanned aerial vehicles (UAVs) have experienced a great leap forward over the last decade. It is not surprising anymore to see a UAV being used to accomplish a certain task, which was previously carried out by humans or a former technology. The proliferation of special vision sensors, such as depth cameras, lidar sensors and thermal cameras, and major breakthroughs in computer vision and machine learning fields accelerated the advance of UAV research and technology. However, due to certain unique challenges imposed by UAVs, such as limited payload capacity, unreliable communication link with the ground stations and data safety, UAVs are compelled to perform many tasks on their onboard embedded processing units, which makes it difficult to readily implement the most advanced algorithms on UAVs. This thesis focuses on computer vision and machine learning applications for UAVs equipped with onboard embedded platforms, and presents algorithms that utilize data from multiple modalities. The presented work covers a broad spectrum of algorithms and applications for UAVs, such as indoor UAV perception, 3D understanding with deep learning, UAV localization, and structural inspection with UAVs.

Visual guidance and scene understanding without relying on pre-installed tags or markers is the desired approach for fully autonomous navigation of UAVs in conjunction with the global positioning systems (GPS), or especially when GPS information is either unavailable or unreliable. Thus, semantic and geometric understanding of the surroundings become vital to utilize vision as guidance in the autonomous navigation pipelines. In this context, first, robust altitude measurement, safe landing zone detection and doorway detection methods are presented for autonomous UAVs operating indoors. These approaches are implemented on Google Project Tango platform, which is an embedded platform equipped with various sensors including a depth camera. Next, a modified capsule network for 3D object classification is presented with weight optimization so that the network can be fit and run on memory-constrained platforms. Then, a semantic segmentation method for 3D point clouds is developed for a more general visual perception on a UAV equipped with a 3D vision sensor.

Next, this thesis presents algorithms for structural health monitoring applications involving UAVs. First, a 3D point cloud-based, drift-free and lightweight localization method is presented for depth camera-equipped UAVs that perform bridge inspection, where GPS signal is unreliable. Next, a thermal leakage detection algorithm is presented for detecting thermal anomalies on building envelopes using aerial thermography from UAVs. Then, building on our thermal anomaly identification expertise gained on the previous task, a novel performance anomaly identification metric (AIM) is presented for more reliable performance evaluation of thermal anomaly identification methods.


Open Access