Computer Science Events
[PAST EVENT] Yichen Luo, Computer Science - Thesis Defense
Location
Chancellors Hall (formerly Tyler Hall), small seminar300 James Blair Dr
Williamsburg, VA 23185Map this location
Abstract:
Software-Defined Vehicles (SDVs) are equipped with a variety of sensors, with cameras playing an important role in enabling intelligent and safety-critical applications. However, these systems face significant challenges due to the diverse visual disturbances introduced by SDV mobility, including inclement weather and unexpected occlusions, which contribute substantial noise to sensor data. Unlike static camera systems, SDV cameras are subject to rapidly changing noise patterns and motion blur resulting from vehicle movement. Despite these challenges, prior research in camera-based perception has overlooked the influence of vehicle speed. To bridge this gap, this paper presents a comprehensive quantitative analysis of in-transit visual noise, leveraging a public dataset, the CARLA simulator, a robotic vehicle, and real-world driving data. Furthermore, we introduce TransitNet, a novel model specifically designed to mitigate in-transit camera noise and improve perception performance at higher speeds. We also provide in-depth observations and insights drawn from extensive experimental evaluations.
Sponsored by: Computer Science