Lane Detection using Classical Computer Vision

This project implements a complete lane detection pipeline using OpenCV and traditional computer vision techniques.
It demonstrates the end-to-end process of detecting road lane markings from an image using noise filtering, edge detection, Hough Transform, and RANSAC line fitting — without using deep learning models.

---

🧠 Overview

• Loaded a road image and converted it to grayscale.
• Added and analyzed Gaussian and Salt & Pepper noise for robustness testing.
• Applied multiple denoising filters (Gaussian, Median, Bilateral) and evaluated their performance using PSNR.
• Detected edges automatically using a custom Auto-Canny function based on image statistics.
• Used Hough Line Transform and RANSAC algorithms to extract dominant lane lines.
• Compared both methods quantitatively using coverage metrics and visual overlays.

---

⚙️ Techniques Used

Stage	Technique	Purpose
Preprocessing	Grayscale Conversion	Simplify computation
Noise Simulation	Gaussian / Salt & Pepper	Evaluate denoising methods
Denoising	Gaussian / Median / Bilateral filters	Remove image noise
Edge Detection	Auto-Canny	Extract lane boundaries
Line Detection	HoughLinesP & RANSAC	Estimate lane lines
Evaluation	PSNR & Line Coverage	Quantify filtering and line quality

---

🧩 Pipeline Summary

1. Load and preprocess image

   • Convert to grayscale
   • Add noise for evaluation

2. Apply filters

   • Gaussian blur
   • Median blur
   • Bilateral filtering

3. Measure PSNR

   • Quantify denoising performance

4. Edge Detection

   • Use auto-threshold Canny edge detector

5. Line Detection

   • Detect lines with Hough Transform
   • Fit lane lines with RANSAC (left/right lanes separately)

6. Compare methods

   • Calculate coverage between detected lines and actual edges
   • Overlay RANSAC (red) vs Hough (green) on the image

---

📊 Example Results

Method	Description	Output
Gaussian Denoising	Smooths random noise
Auto-Canny	Adaptive edge detection
Hough vs RANSAC	Line detection comparison

RANSAC (red) offers cleaner, more stable lane lines than Hough (green), especially in noisy or uneven lighting conditions.

---

🧰 Requirements

pip install opencv-python matplotlib numpy

Evaluation Metrics

PSNR (Peak Signal-to-Noise Ratio): Measures image quality after denoising.

Line Coverage (%): Percentage of edge pixels covered by detected lines (for both Hough and RANSAC).

🚀 Key Functions:

add_gaussian_noise() / add_salt_pepper() – Adds noise for testing

auto_canny() – Automatically selects Canny thresholds

detect_lines_houghp() – Detects lines using Hough transform

ransac_line_fit() / ransac_side() – Fits left/right lane lines robustly

line_coverage() – Quantifies detection quality

📚 Concepts Demonstrated:

Noise modeling and denoising evaluation

Adaptive edge detection

Line detection using Hough and RANSAC

Quantitative evaluation with PSNR and coverage metrics