3D Road Lane Classification with Improved Texture Patterns and Optimized Deep Classifier

Area of Science:

• Computer Vision
• Machine Learning
• Autonomous Systems

Background:

• Accurate road and lane understanding is crucial for autonomous driving, yet current perceptual methods face limitations.
• 3D lane detection, estimating precise drivable lane positions, is a key research area for autonomous vehicles.

Purpose of the Study:

• To propose a novel two-phase technique for 3D lane detection using 3D images.
• To improve road/non-road and lane/non-lane classification accuracy.

Main Methods:

• Phase I: Road/non-road classification using local texton XOR pattern (LTXOR), local Gabor binary pattern histogram sequence (LGBPHS), median ternary pattern (MTP) features with bidirectional gated recurrent unit (BI-GRU).
• Phase II: Lane/non-lane classification using similar features with an optimized BI-GRU, where weights are optimized via self-improved honey badger optimization (SI-HBO).

Main Results:

• The proposed BI-GRU + SI-HBO achieved a precision of 0.946 (db 1).
• The best-case accuracy for BI-GRU + SI-HBO reached 0.928, outperforming standard honey badger optimization.
• The self-improved honey badger optimization (SI-HBO) demonstrated superior performance compared to other optimization methods.

Conclusions:

• The developed two-phase approach effectively enhances 3D lane detection capabilities for autonomous vehicles.
• The integration of BI-GRU with SI-HBO offers a promising direction for improving the accuracy and robustness of lane detection systems.