This paper describes a high level strategy to guarantee the revisit of a target in underwater environment by an Autonomous Underwater Vehicle (AUV) subject to high drift in position and equipped with specific exteroceptive sensors such as sonars or cameras. It assumes the availability of a certain a priori map of the environment. Firstly, based on the Field Of View (FOV) of the exteroceptive sensor and the characteristics of the different landmarks in the map, a registration map is computed. This indicates the sets of the robot configurations able to detect the different landmark considered. All these sets called relocation areas enable to relocate the robot position and reduce its uncertainty by detecting the geolocalized landmarks. These relocation areas are the nodes of a hyper-graph. Between two nodes, the vehicle navigates by dead-reckoning. Based on the backward reach set of a registration map, the links between the different registration maps can be created according to a motion model taking into account uncertainties. Finally, a backward graph search minimizes a cost function to propose a strategy of revisit of intermediate landmarks to connect the starting position and the relocation area(s) of the target you would like to revisit.