This paper presents a self-calibration technique for a camera mounted on an underwater vehicle designed to perform the 3D reconstruction of underwater scenes. Our aim is to identify the intrinsic parameters of the camera with methods that are adapted to the operational constraints on Ifremer's underwater vehicles. The optical system is composed by a single vertical camera located below the underwater vehicle and looking downwards. The motion of the vehicle can be measured through navigation sensors and the observed 3D scene is always unknown. The use of a moving camera is not an obstacle for the application of stereoscopic methods. Nevertheless, the camera motion enables the use of robust algorithms for points matching, but impoverishes perspective effects between several images. Therefore, we are interested in the analysis of the conditions in which the procedure of self-calibration is valid and reliable, i.e.: the 3D characteristics of the scene and the camera motion. This paper presents the steps necessary for the camera self-calibration in an underwater environment: the extraction and the tracking of features in several successive images, the fundamental matrix estimation and the intrinsic parameters identification. Several tests and results are presented.