tomazas / opencv-lane-vehicle-track

OpenCV implementation of lane and vehicle tracking

###Project implements a basic realtime lane and vehicle tracking using OpenCV.###

Screenshots:

Implemented with:

• OpenCV 2.3
• C/C++ using Microsoft Visual Studio 2010 IDE.

OpenCV features used & used techniques:

• Gaussian smoothing for image noise removal
• Canny edge detection [1]
• Hough transform for line detection
• Haar features for vehicle detection (hypothesis generation) [2]
• Vehicle hypothesis verification using horizontal edges and symmetry [3]

Possible improvements:

• k-Means clustering for Hough lines
• Kalman/Gabor/RANSAC filtering of sampled data
• KLT (Kanade-Lucas-Tomasi) feature tracker for vehicle tracking
• Vanishing point detection using Gaussian probability model
• Better lane tracking(probability methods), stability & accuracy
• More accurate vehicle hypothesis checking
• Alternative IPM(inverse perspective mapping) lane detection method
• Road area extraction & detection for roads without lanes
• Ability to process night vision situations
• Include road shadow removal
• Speed upgrades
• Road sign and traffic lights detection

References:

1. Canny, J., "A Computational Approach To Edge Detection", IEEE Trans. Pattern Analysis and Machine Intelligence, 1986
2. Viola and Jones, "Rapid object detection using a boosted cascade of simple features", Computer Vision and Pattern Recognition, 2001
3. King Hann Lim et al. "Lane-Vehicle Detection and Tracking", IMECS, 2009

#####Training data used from: ##### California Institute of Technology SURF project

Projects using our source code base & samples (YouTube video):

Example project homepage