Camera-Based Terrain Surface Recognition

Alain Dunoyer (Jaguar Landrover)

The aim of the project is to assess and develop methods for automatically
detecting different terrain surfaces based on colour camera images. In order
to minimize the complexity of such a system the scope will be limited to the
detection of snow, grass and tarmac.

A challenge will be not to rely on the colour cue for the detection system.
This is primarily due to the fact that colour stability cannot be guaranteed
with automotive cameras, especially at night or under poor lighting
conditions. The methods may therefore need to concentrate on pattern
recognition rather than colour histogram based methods.

Also, one of the outcome of this project should be to clarify if the better
approach should be to concentrate the analysis on a subsection of the image
(primarily where the vehicle is about to be driven) or if the entire image
(contextual analysis) should be considered.

For this project JLR will provide real-time camera images for the analysis
phase and will also offer the opportunity to validate any developed
methodology directly into a vehicle if the implementation is feasible within
the project timescale.