The purpose of this work is to detect the obstacles on the road thanks to a camera mounted on a vehicle. Due to fixed constraints, an overview of various existing methods has shown that only motion analysis in the image sequence can solve this problem. Indeed, this method must permit to detect all types of obstacles on all kinds of roads with only one camera. For that, we have developed a method of parametric motion estimation by wavelet analysis. This multi- resolution method enabled us to overcome the aliasing problem. Then, we have proposed to solve the motion detection problem in a scene filmed with a mobile camera using a hierarchical Markovian mo- deling deduced in a natural way of the multi-resolution motion estimation. Next, we have introduced a motion segmentation method between two images without a priori knowledge and without hypothesis of dominant motion thanks to a successive refinement of the segmentation from coarse scale to fine scale of the image. All methods (estimation, detection and segmentation) have been validated in experiments on synthetic and real sequences. Lastly, those were adapted to the concrete problem aimed by this thesis : the obstacle detection in road sequences. The use of wavelet and hierarchical Markovian model leads to inexpensive solutions in computing times.
