Event
Ph.D. Research Proposal Exam: Ruiqi Xian
Friday, January 10, 2025
12:00 p.m.-2:00 p.m.
umd.zoom.us/my/dmanocha
Maria Hoo
301 405 3681
mch@umd.edu
ANNOUNCEMENT: Ph.D. Research Proposal Exam
Name: Ruiqi Xian
Committee:
Professor Dinesh Manocha (Chair)
Professor Shuvra Bhattacharyya
Professor Pratap Tokekar
Date/time: Friday, January 10, 2025 at 12pm-2pm EST
Location: umd.zoom.us/my/dmanocha
Title: Elevating Aerial Perception: From Observation to Understanding in Aerial Videos
Abstract:
Understanding and analyzing aerial videos captured from above or oblique angles, including those from Unmanned Aerial Vehicles (UAVs), surveillance cameras, and satellites, is critical for applications such as security, environmental monitoring, and disaster response. These videos provide unique perspectives on dynamic environments, but they also present significant challenges due to factors like occlusion, dynamic backgrounds, and viewpoint variability. My research focuses on developing innovative methods and frameworks to address these challenges, advancing aerial perception across diverse real-world scenarios.
Our work mainly focuses on aerial video action recognition, spanning three key areas. First, we develop edge computing solutions to create algorithms and recognition methods optimized for resource-constrained platforms such as drones and other edge devices. Second, we introduce information theory-guided approaches for robust feature alignment and adaptive frame selection, mitigating the effects of dynamic backgrounds and camera motions while capturing the most informative frames. Third, we leverage self-supervised learning techniques to enhance aerial video pre-training, incorporating object-level information to accelerate training, reduce memory usage, and improve downstream recognition performance. Beyond action recognition, we also contribute to cross-modal global localization by designing methods that address scale variations and representation gaps between LiDAR and satellite maps, and to multi-camera multi-object tracking from surveillance cameras through self-supervised approaches that reduce reliance on manual annotations and camera calibration. By addressing the challenges in aerial data, my research provides scalable and efficient solutions that enhance aerial video understanding, enabling a wide range of applications in surveillance, autonomous systems, and global monitoring.
