Brigit Schroeder
May 3, 2012

Line Following and Object Detection Through A Birds Eye View


The project involves a method of robot line following and obstacle detection for my final project through the use of image understanding. Image understanding is the process of interpreting regions in an image after they have been segmented and detected by a sequence of image processing techniques. For line following, within the scope of the IGVC, identifying and understanding which regions are white painted lines, grassy areas and basic obstacles such as traffic barrels, is critical. Once the lines other obstacles have been detected in imagery, the physical locations of the lines need to be derived (or at best, approximated) in order to allow the robot to steer away and stay on course, which can be done using a bird's eye projection calibrated to a physical metric coordinate system in front of the robot.

IGVC Course

Line and Obstacle Detection Projected To A Bird's Eye View From Sample IGVC Video

Simulated Course

Improved Line and Obstacle Detection Projected To A Bird's Eye View From Simulated IGVC Course

Concepts Demonstrated

  • Analysis of HSV (hue, saturation and value) color space is used to create multiple binary masks which combined and filtered to produce a final "line and barrel" mask.
  • Projective geometry and calibration targets are used to warp the robot's perspective view into a "bird's eye view" (top-down). This projection remove the perspective

from the original image.

  • Connected components analysis is used to segment and differentiate lines from barrels and also remove unwanted noise from the image.


The innovation this project involves a create and observant usage of the HSV color spaceand also using the geometric properties of various targets to isolate them. The "bird's eye" perspective has also been transformed into a "in front of the robot" view which would be the eventual input into an occupancy grid.

Process Block Diagram