A Fast Learning Method for Accurate and Robust Lane Detection Using Two-Stage Feature Extraction with YOLO v3

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Accurate lane detection is crucial for advanced driver-assistance systems (ADAS) and autonomous driving.
• Existing lane detection methods struggle with complex scenarios like varying illumination and occlusions.
• The YOLO v3 (You Only Look Once, v3) algorithm offers a foundation but requires adaptation for precise lane feature learning.

Purpose of the Study:

• To develop an adaptive lane feature learning algorithm for robust lane detection in diverse and challenging driving conditions.
• To enhance the accuracy and efficiency of lane detection systems through a novel two-stage learning approach.
• To improve the reliability of autonomous systems by refining their ability to identify lane markings.

Main Methods:

• A two-stage learning network was constructed, modifying the YOLO v3 architecture for lane detection.
• An automatic lane label image generation method was proposed to improve training efficiency for the first-stage network.
• An adaptive edge detection algorithm using the Canny operator was employed for lane relocation, with unrecognized lanes being shielded.

Main Results:

• Experiments on the KITTI and Caltech datasets demonstrated high accuracy and speed for the proposed second-stage model.
• The adaptive feature learning algorithm effectively handled complex scenarios, outperforming standard methods.
• The two-stage approach, incorporating automatic labeling and edge-based relocation, significantly improved detection performance.

Conclusions:

• The proposed adaptive lane feature learning algorithm significantly enhances lane detection accuracy and speed in complex scenarios.
• The two-stage YOLO v3-based network, combined with adaptive edge detection, provides a robust solution for real-world lane identification.
• This research contributes to the advancement of safer and more reliable autonomous driving technologies.