Lane Detection Segmentation using YOLOv11

This repository provides a deep learning solution for lane detection segmentation using the YOLOv11 segmentation model. The project is implemented in Python using the Ultralytics YOLO framework and is optimized for training on custom lane datasets.

---

🔹 Project Overview

Lane detection is a crucial task in autonomous driving and advanced driver-assistance systems (ADAS). This project uses YOLOv11-seg to accurately segment lane markings, including solid and dashed lines on both left and right boundaries.

Key Features:

• Segment lanes in real-time
• Handles multiple lane types (solid/dashed)
• High precision and recall for both bounding boxes and masks
• Early stopping to prevent overfitting

---

🔹 Model Training

The model is trained using YOLOv11-seg with the following configuration:

from ultralytics import YOLO

model = YOLO('yolo11n-seg.pt')
!yolo segment train \
  data="data.yaml" \
  model=yolo11n-seg.pt \
  epochs=200 \
  patience=30 \
  imgsz=640 \
  name="lane-seg"

Training Details:

• Epochs: 200 (training stopped early at 144 due to EarlyStopping)
• Image size: 640
• Patience: 30 epochs
• Best Model: best.pt
• Hardware: Tesla T4 GPU, CUDA acceleration

---

🔹 Performance Metrics

Metric	Value
Box Precision	0.889
Box Recall	0.855
Box mAP50	0.908
Box mAP50-95	0.747
Mask Precision	0.882
Mask Recall	0.832
Mask mAP50	0.889
Mask mAP50-95	0.559

Per Class Performance:

Lane Type	Images	Instances	Box(P)	Box(R)	mAP50	Mask(P)	Mask(R)	Mask mAP50
Left Boundary - Dashed	158	356	0.935	0.762	0.888	0.931	0.742	0.877
Left Boundary - Solid	138	139	0.932	0.942	0.952	0.938	0.942	0.956
Right Boundary - Dashed	86	172	0.862	0.779	0.847	0.840	0.730	0.796
Right Boundary - Solid	172	174	0.826	0.937	0.944	0.914	0.926	0.613

---

🔹 Validation and Inference

After training, the best model best.pt was validated on a test set, showing high accuracy and mask quality.

Inference Speed:

• Preprocess: 0.1 ms per image
• Inference: 2.7 ms per image
• Postprocess: 4.3 ms per image

The trained model can be used to segment lane markings on new images with high reliability.

results = model("/path/to/test/images", conf=0.6)
results.show()

---

🔹 Key Highlights

• Uses YOLOv11-seg, a lightweight yet powerful segmentation model
• EarlyStopping prevents overfitting and saves the best weights automatically
• Supports multiple lane types with high precision and recall
• Can be integrated into autonomous driving pipelines or lane monitoring systems

---

🔹 Future Improvements

• Train on a larger, more diverse dataset for improved generalization
• Deploy the model on edge devices for real-time lane detection
• Combine with object detection for a complete ADAS solution

---