tum-phoenix / drive_ros_image_recognition

image recognition (lanes, crossings, start lines, ....)

MIT License

2 stars 1 forks source link

[LDCA/LDML] Lane Detection #13

Open a-0-dev opened 5 years ago

a-0-dev commented 5 years ago

Objective: Lane Detection for our CC20 Car

The lane Detection is an important task in autonomous driving.

Goal

Stable and fast line detection for trajectory generation
detect the current lane and identify the different lines and line types (outer left, middle, double, solid, dashed)
Robustness against missing line parts (Broken lines are part of the cup)
Robustness against obstacles and ramp
handling for curves
handling of crossings and intersections with priority lines.
Pedestrian island and zebra crossings

Setup:

Our previous line detection can be found: https://github.com/tum-phoenix/drive_ros_image_recognition/tree/polynom You can try different computer vision approaches or machine learning approaches.

Machine learning - line detection For the machine learning approach there is also a different issue: https://github.com/tum-phoenix/drive_ml/issues/19
Hough lines
Road segmentation
...

A template with basic functions can be found under: https://github.com/tum-phoenix/drive_ml/issues/19

Tasks:

[ ] Implement one or more line detection approaches (for comparison)
[ ] Focus on line detection and trajectory at first and check the stability of the trajectory
[ ] Extract polynomial for the trajectory and the environment model
[ ] Detect the current lane and identify the different lines and line types
[ ] Detect Crossings and special street parts, like islands
[ ] Estimate the curvation radius
[ ] Compare different approaches

xXBasti commented 5 years ago

Maybe that can be helpful: https://github.com/tatsuyah/Lane-Lines-Detection-Python-OpenCV