For human beings, using their eyes for navigating streets, sidewalks or even grassland is an everyday process that does not require much thought. It seems to be straightforward to let robots do the same thing, using their built-in cameras.
However, while most mobile robots do have at least one camera installed, they are not really using it, apart from providing images to the human operator. The reason for this is that processing images in order to gain knowledge about the robot's surroundings is a complex and error-prone operation.
The Path Detection module analyzes color video images and determines the properties of the path in front of the robot, based on aspects such as color or texture. These properties are then inserted into a model of the world, and then analyzed with statistical methods. The result is a best estimate about which parts of the image are part of the road, and which ones are not (and should therefore be deemed undriveable).
Fig. 1: Result of path detection analysis
This estimate can then directly be used to steer a vehicle along the recognized driveable path, or just taken into consideration when performing path planning by other means. A powerful software application can be used to perform manual analysis and optimization. This application allows developers to understand and analyze, as well as optimize and customize the path detection.
Fig. 2: The RoadVision software contains powerful analysis facilities