Adaptive methods in image sequence analysis:
- Visual motion detection/estimation
In case of a stationary camera pixel based motion detection schemas work already very well. In following examples the detected visual motion corresponds directly to moving objects, although small camera movements (as in the right sequence) already introduce improper visual motion of stationary background (these are some results of my former project ERBOB at FORWISS Erlangen):
MPEG.
MPEG.
[go back to top],
- Road following
Consider the approach to road recognition and road following described in our paper published in the Int. Journal of Computer Vision, 1998
[go back to top],
- Object tracking and recognition
In case of a moving camera, with partially unknown camera movement, in scenes with many moving objects a pixel based motion detection is usually not reliable enough. The visual motion does not correspond directly to a moving object. For a robust object motion estimation a model-based processing scheme can be applied. The stabilization of object hypotheses and consistency evaluation of multiple hypotheses is performed by means of adaptive methods. My approach to object detection and tracking for car driver assistance has been published in several papers, for example at IAPR Workshop MVA 1996 or in the MG&V journal.
Here is an example how a simple 3-D object model can already lead to proper moving object detection and true motion estimation:
MPEG movie.
- Road following
[go back to top],