Improvement of Automatic Man-made Object Detection in Underwater Videos by Using of Navigational Information

In order to detect man-made objects in underwater video, we propose and validate a novel method based on background subtraction methods. The other main contribution is the introduction of a priori elements deduced from positioning sensors. These elements allow to enhance the visibility of underwater objects thanks to the calculation of the sun position in relation to the vehicle position, the detection with the distance from the object and the post-processing with constraints defined on the vehicle movements. These constraints allow to reject false detections and to better know the position of the detected object. We tested our algorithm on data acquired at sea and show that we improve detection results and decrease false alarm rate, comparing to our former work. Both algorithms have been applied on the same videos. We still have to increase the true detection rate while reducing processing time i.e. processing time should be close to video rate.