Applying Visual Processing to GPS Mapping of Trackside Structures

This paper describes an application of pointwise structure from known motion to assist in the construction of roadside and trackside maps. A single camera is used to record the view from the front of the vehicle, and positional information from an onboard GPS receiver is used to provide the location and orientation of the vehicle. After calibration from known structures, the transformation from world coordinates to image coordinates can be specified for each frame, and multiple views of the same feature, allow the world coordinates of the features to be recovered. Objects must be selected and labelled by hand, but thereafter tracking, recovery and mapping are automated. Examples are given of the recovery and logging of structures along a railway line.