The planetary rovers must be able to accurately, quickly, and reliably
perceive the terrain for navigation. The main objective of the thesis is to
realize a laser-based optical system for obstacle detection at unknown
terrain for a planetary rover. The optical system consists of an active
light emitter that projects a pattern of laser beams onto terrain surface,
and a CMOS camera captures images and the image processing software
identifies the variations in the pattern as it encounters differen
obstacles.
The proposed Active Light system is deployed on a MERLIN rover. The approach
based on active light is expected to offer substantially reduced sensing
costs, allowing more reliable navigation through unstructured areas.
The implemented purely vision based algorithms rely on capturing and
detecting deformation in the laser pattern. Simple straight forward
Template matching method and Boundary masking method are proposed
initially. Subsequently optical flow [28] based and CamShift [29] based
obstacle detection methods are implemented to overcome the limitations
posed by earlier methods. The real world environment is highly unstructured
and dynamic, each of the four algorithms, for obstacle detection, have
certain advantages and limitations. The dynamics in the captured image due
to rover motion is unpredictable, since it is governed by an unknown number
of parameters. Obstacle detection becomes a more complicated problem in
outdoor scenes, where numerous factors are controlling the scene in a very
unpredictable way. A Meanshift [30] based approach is used for estimating
pattern shifts in video sequence due to rover vibrations. Active Light
sensor reliably recognizes obstacles irrespective of their shape in real-
time. The performance of the system is evaluated on wide variety of
terrains, for each of the image processing based obstacle dete...