Kalman Filter based Depth from Motion with Fast Convergence
Uwe Franke, Clemens Rabe
The extraction of depth is a prerequisite for many applications in robotics and driver assistance. Examples are obstacle detection, collision avoidance, and parking. This paper presents a new Kalman filter based depth from motion approach. Thanks to multiple filters running in parallel the rate of convergence is significantly higher than in direct methods, especially if the vehicle drives slowly. A goodness-of-fit test fuses the states of the different filters in an optimum manner. In addition, this test allows to distinguish between static and moving obstacles.