Kfir Twizer- CW Radar-Based Road Environment Detection and Matching for Robust Localization in GNSS-denied scenarios

Electrical Engineering Systems Seminar

Speaker: Kfir Twizer

M.Sc. student under the supervision of Prof. Ben-Zion Bobrovsky

Sunday, 28^th January 2024, at 15:00

Room 011, Kitot Building, Faculty of Engineering

FMCW Radar-Based Road Environment Detection and Matching for Robust Localization in GNSS-denied scenarios

Abstract

Real-time global localization information is a critical component in modern navigation and perception systems, as it enables effective navigation, route planning and environment awareness. During GNSS outages, or under poor signal conditions, other complementary sources are employed. State-of-the-art techniques often use camera and/or LiDAR sensors to perform this task. However, these sensors performance is vulnerable to adverse weather conditions like rain, fog or snow. In such scenarios, Radars emerge as reliable primary sensors, aligning with the redundancy requirements of the automotive industry.

In this work, we propose a robust and efficient model for radar-based self-localization in urban roads. Our model extract relevant information for navigation from radar measurements; stationary obstacles along-side roads are detected and tracked through an enhanced version of extended target tracking framework developed in this work; Road users' movement in the range of interest is detected as well. This information is then matched to different classes of HD semantic maps using an innovative map-matching algorithm developed in this study, which integrates Likelihood Fields. Ego pose is being estimated over time according to the map-matching score.

Our model eliminates the need for pre-generated custom occupancy grid maps, known for their maintenance challenges, non-availability and storage costs. Instead, it seamlessly integrates with widely used semantic maps available today. Across diverse urban scenarios from the nuScenes public dataset, our model demonstrates a 1m/1.5m RMS lateral/longitudinal error correspondingly during typical periods of GNSS outage.

השתתפות בסמינר תיתן קרדיט שמיעה = עפ"י רישום שם מלא + מספר ת.ז. בדף הנוכחות שיועבר באולם במהלך הסמינר