This paper presents a method to build ne resolution digital terrain maps, on the basis of a set of low altitude aerial stereovision images. Stereovision images are the only data available: to build a global map, we introduce an algorithm that uses interest point matches between successive images to estimate their relative position. Experimental results are presented and discussed, they show how one can build thousands square meters ne resolution maps, integrating hundreds of stereovision images acquired by a tethered blimp.