报告人：Zimin Xia
主持人：Jingyuan Yang
日期：2024.06.24
时间：4:00pm
地点：深圳大学致真楼801
Abstract
Vehicle self-localization is an essential task in autonomous driving, typically relying on High-Definition (HD) maps and LiDAR sensors. However, creating and maintaining up-to-date HD maps is costly and labor-intensive, and LiDAR sensors are not widely available in consumer vehicles. Therefore, exploring alternative map sources for camera-based localization is both important and practical. Aerial images, with their rich appearance information and public accessibility, serve as a promising map source.
This talk will delve into a novel vehicle localization technique that utilizes only images taken by a vehicle-mounted camera and geo-referenced aerial images, known as ground-to-aerial cross-view visual localization. The presentation will introduce the relevance of this task, provide an overview of state-of-the-art methodologies, identify key challenges, and suggest potential future research directions.
Bio
Zimin Xia is a PostDoc researcher at the Visual Intelligence for Transportation Lab at École Polytechnique Fédérale de Lausanne (EPFL), advised by Prof. Alexandre Alahi. He completed his Ph.D. in the Intelligent Vehicles group at Delft University of Technology, under the supervision of Prof. Julian Kooij and Prof. Dariu Gavrila. His Ph.D. project was funded by the Dutch Research Council and the digital mapping company TomTom. Prior to his Ph.D., Zimin completed his MSc in Geomatics at the University of Stuttgart, Germany. He received his BSc degree in Geomatics Engineering from Wuhan University, China.
Zimin's research interests lie at the intersection of computer vision and mobile robotics. He has been focusing on the visual localization of autonomous vehicles, specifically ground-to-aerial cross-view localization. His work has been published in top computer vision and robotics venues, including T-PAMI, CVPR, ECCV, and RA-L.
