FF-HPINet: A Flipped Feature and Hierarchical Position Information Extraction Network for Lane Detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Lane detection is crucial for autonomous driving systems.
• Current deep learning models have limitations in utilizing visual symmetry and position information for lane detection.

Purpose of the Study:

• To propose FF-HPINet, a novel deep learning approach for enhanced lane detection.
• To address the underutilization of visual symmetry and position information in existing methods.

Main Methods:

• Introduced a Flipped Feature Extraction module to capture symmetrical features and semantic information from varied receptive fields.
• Developed a Hierarchical Position Information Extraction module to precisely mine lane position data.
• Incorporated a Deformable Context Extraction module to focus on relevant foreground elements and contextual details.

Main Results:

• Achieved a 97.00% F1 score on the TuSimple dataset.
• Attained a 76.84% F1 score on the CULane dataset, demonstrating superior performance.

Conclusions:

• FF-HPINet effectively leverages feature symmetry and hierarchical position information for improved lane detection.
• The proposed modules enhance precision and accuracy in autonomous driving lane identification systems.