Sapir Kontente-CLRmatchNet: Enhancing Curved Lane Detection with Deep Matching Process

Electrical Engineering Systems Seminar

Speaker: Sapir Kontente

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

Wednesday, 25^th February 2024, at 15:00

Room 011, Kitot Building, Faculty of Engineering

CLRmatchNet: Enhancing Curved Lane Detection with Deep Matching Process

Abstract

Lane detection is crucial for safe autonomous navigation, providing essential data for accurate vehicle guidance. Currently, anchor-based detectors are the leading paradigm for lane detection and are commonly combined with a classical label assignment step during training, labeling model predictions for optimal learning. Accurate label assignment has great impact on the model performance, that is usually relying on a predefined classical cost function evaluating GT-prediction alignment. However, classical label assignment methods face limitations due to their reliance on predefined cost functions derived from low-dimensional models, potentially impacting their optimality.

This study introduces MatchNet, a deep learning submodule aimed at addressing challenges in the label assignment stage for lane detection. Integrated into the state-of-the-art lane detection network, CLRNet, MatchNet replaces the classic label assignment process and exhibits significant enhancements, particularly in detecting curved lanes. It raises confidence levels in positive lanes and introduces dynamic flexibility in lane matching. The integrated model, CLRmatchNet, surpasses CLRNet, showcasing substantial improvements in scenarios involving curved lanes, with enhancements of +2.8% for ResNet34, +2.3% for ResNet101, and +2.96% for DLA34.

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