Tracking technical objects in outdoor environment based on CAD models

Tracking objects under difficult conditions is a challenging topic in many real life and outdoor applications. In particular persistent occlusion of large parts of the model, difficult lighting conditions, and a noisy background result in a failure of many available algorithms. To overcome these difficult problems we significantly extend the existing RAPiD approach, in particular by putting sample control points at random intervals, and providing the possibility for fix points. We show how to automatically extract an edge model that is suitable for tracking from CAD models. The benefit of our approach is presented by means of a real life outdoor application scenario suffering from the mentioned bad conditions. We obtain relative translation and rotation errors that are only moderately higher than those in less challenging setups.