SMPPALD-Segmentation Mask Post-Processing Algorithm for Improved Lane Detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Automotive Engineering

Background:

• Advanced Driver Assistance Systems (ADAS) increasingly rely on accurate Lane Detection (LD).
• Current LD algorithms often use neural networks to generate lane marking segmentation masks.
• Post-processing these masks can potentially improve detection accuracy.

Purpose of the Study:

• To introduce SMPPALD (Segmentation Mask Post-Processing Algorithm for improved Lane Detection), a novel algorithm for enhancing LD.
• To refine neural network-generated segmentation masks into precise lane marking point lists.
• To improve the overall accuracy and robustness of lane detection systems.

Main Methods:

• Developed SMPPALD, a post-processing algorithm applying geometric and contextual rules to segmentation masks.
• Integrated SMPPALD with the Spatial Convolutional Neural Network (SCNN) for lane detection.
• Evaluated performance on three diverse datasets: CULane, TuSimple, and LLAMAS.

Main Results:

• SMPPALD significantly improved the F1 measure compared to SCNN on the TuSimple and LLAMAS datasets.
• The algorithm demonstrated superior performance over SCNN in most categories on the CULane dataset.
• Post-processing segmentation masks led to more accurate lane marking identification.

Conclusions:

• SMPPALD offers a valuable enhancement for neural network-based lane detection systems.
• The proposed post-processing method effectively improves lane detection accuracy in ADAS.
• SMPPALD demonstrates broad applicability across different datasets and network architectures.