Abstract:
Unmanned Ground Vehicles, or Intelligent Vehicles as they are known, are vehicles
that can drive without the need for human interaction. A lot of research work
has been done in this branch of robotics in recent years, the most famous one being
the DARPA challenge. Using sensors such as laser, Global Positioning System ( GPS
) and cameras, intelligent vehicles will process road information and act accordingly.
Therefore, it is vital that the vehicles process information such as tra c signs and
distance traveled as accurately as possible. This thesis addresses two issues regarding
intelligent vehicles: tra c sign recognition and distance estimation. These two issues
are addressed using a monocular camera as a sensor input.
For tra c sign detection and recognition, I used the signs based on those used
in the Thailand Intelligent Vehicle Challenge organized by the Thai Robotics Society(
TRS ). Signs are detected based on their physical properties and template matching is
used for sign recognition.
Visual odometry is used for distance estimation. Visual odometry is essential the
estimation of distance traveled based on feature displacements in two images taken at
di erent points in time. The camera projection matrix P is obtained using Random
Sampling And Consensus( RANSAC ) and distance estimation is calculated using
homography decomposition.