This project implements an advanced lane detection system without using any machine learning models, relying solely on classical computer vision techniques using OpenCV and NumPy. The pipeline processes video frames (or static road images) and detects lane boundaries, overlays them back on the road, and calculates lane curvature and vehicle position relative to the center.
Note: The parameters are based on the images I used for testing. Refer to What to Change for more details.
📹 Check out the lane detection in action below! Click the Link
- Clone the repository:
git clone https://github.com/Haseebasif7/Advance-Lane-Detection.git
cd Advance-Lane-Detection- Install dependencies:
Make sure you have Python installed. Then install the required libraries:
pip install -r requirements.txt- Run the pipeline:
python lane_pipeline/main.pyThe lane detection pipeline involves several key steps to transform raw camera data into accurate lane boundaries and provide useful information for autonomous driving systems. Here's an expanded breakdown of each step in the pipeline:
Purpose: The camera calibration step helps correct any distortion caused by the camera lens. Cameras, especially wide-angle ones, can introduce lens distortion (often called fisheye distortion), which causes straight lines in the real world to appear curved in the image.
How it works:
- We use a set of chessboard images taken from different angles. These images allow us to compute the camera matrix and distortion coefficients, which characterize how the camera distorts the images.
- With these parameters, we can undo the distortion, so the road and lane lines appear as they are in the real world, not warped by lens effects.
This correction is crucial for precise lane detection, as any distortion can lead to inaccurate lane marking detection and misinterpretation of road boundaries.
Purpose: After computing the camera matrix and distortion coefficients, we apply a distortion correction to the images.
How it works:
- Using the calibration parameters from the previous step, we apply an algorithm (typically
cv2.undistort()in OpenCV) that removes the distortion in each input image. - The goal is to produce an image where straight lines (like lane boundaries) are perfectly straight, helping improve the accuracy of subsequent steps in the pipeline.
This step ensures that the raw data you work with is geometrically accurate, without any optical errors introduced by the camera hardware.
Purpose: The perspective transform is used to transform the camera’s view of the road into a top-down (bird's-eye view) perspective. This simplifies lane detection and makes the road appear as if viewed from above, where lanes appear parallel and more consistent in shape and width.
How it works:
- We define a region of interest (ROI) in the image that contains the road and lanes. This is the area where lane detection will occur.
- Using a perspective transform, we map the original camera image to a top-down view using a homography matrix. This matrix defines the transformation between the original camera view and the bird's-eye view.
- The result is an image that simplifies lane detection, where the lanes appear as straight, parallel lines, making it easier to detect lane boundaries and calculate curvature.
This step mimics how human drivers see the road from above, giving us a clearer view of the lane markings.
Purpose: In this step, we convert the image into a binary format where lane line pixels are highlighted, and the rest of the image is black. This simplifies the task of detecting lanes by focusing on relevant features.
How it works:
- We use thresholds (such as R-Binaary) to highlight the lane lines, based on their distinctive colors in the road image.
- After applying these techniques, we combine the results into a binary image where the lane pixels are white, and the rest of the image is black.
Purpose: The goal of this step is to identify the pixels that belong to the left and right lane boundaries, and fit a curve to those points for accurate lane detection.
How it works:
- We first create a histogram of the pixel positions along the bottom of the image to locate the approximate position of the lane lines.
- Using a sliding window technique, we search for the lane pixels in a defined region, iterating upwards from the bottom of the image.
- Once we find the pixels belonging to the left and right lanes, we fit polynomial curves (typically quadratic) to these points. This results in a precise representation of the lane boundaries, even on curved roads.
Purpose: After detecting the lane boundaries, we calculate the curvature of the lanes and the vehicle's position relative to the center of the lane. These metrics are essential for understanding how sharp the turn is and where the vehicle is positioned on the road.
How it works:
- Radius of curvature is calculated using the equation that fits a second-order polynomial to the lane points. This gives us the turn radius, which helps us understand how sharp the curve is.
- The vehicle offset is determined by measuring how far the vehicle is from the center of the detected lane. This is calculated in real-world units (e.g., meters) by converting from pixels using a camera calibration matrix.
- These metrics are critical for ADAS systems, allowing the vehicle to adjust its behavior based on the road curvature and position.
Each of these steps is essential in building a robust lane detection system that can perform well in various road conditions and scenarios. By applying these techniques, we can achieve accurate lane detection without relying on complex machine learning models.
Finally, the detected lane lines are warped back to the original image perspective, and both visual and numerical information is overlayed on the frame.