Underwater navigation by video sequence analysis

A method to estimate the trajectory of a submersible vehicle from conventional video images is presented. A movement estimation approach establishes correspondences between the feature points of several selected areas in two successive images. The generalized Hough transform (GHT) is used to estimate the vehicle´s trajectory by determining the translation associated to the selected feature points in the images. The performances of the GHT are improved by using a Kalman filter, which predicts the feature point positions on the next image. Three confidence factors were computed to evaluate the GHT performances in noisy images, which are also used to weight the displacement of each area in the computation of the image mean displacement. The algorithm was tested with several undersea video sequences from an experiment carried out by the French Institute for the Research and Exploitation of the Sea. This approach leads to a robust and accurate trajectory estimation