Hello,
Problem - I need to design a navigation algorithm using Image processing (possibly without using stereo vision) where the robot should navigate between two colored lanes (black and yellow) without crossing any of the lanes. The lanes can be curved at places and straight rest of the times. I am writing down the specifications of the problem and possible solution that I can think of at this moment, if somebody has already worked on similar project please discuss the solution.
Requirements -
1. The solution should give estimation of distance from the camera to the colored lane to avoid crossing the lane
2. The turning direction of the lane curve must be detected to turn the robot in specified direction
Possible solution (That I can think of currently, not sure whether I am correct or not) -
1. Segmenting the colored lanes - this can be done by edge detection or using color histograms to get features.
2. Using Lucas Kanade method for Optical flow we can get direction of pixels (point in the lane) which will represent the tangent at a point on the lane curve, a collective measure then can give the turning direction of the lane.
I found a Youtube video which shows exact result that I am expecting, unfortunately there is no supporting description of method used to get the output. Please see the video to understand the problem.
Thanks and Regards,
Mrityunjay
http://mrityunjay.in/