We present a new approach for localizing a ground target from a UAV using 3D terrain engine. We use inaccurate auxiliary sensors on the UAV to obtain an approximate measurement of the camera pose, we use this pose to move the virtual camera inside the engine, and then automatically we find multiple matches between the two images to find the 3D coordinates of the matches using 3D terrain engine. Finally, we test the coplanarity of the 3D points under the camera, depending on this test, we use coplanar or non-coplanar algorithm to estimate accurate global camera pose. The accurate pose is used for localizing targets seen in the image by transmitting a virtual ray from a pixel according to the camera pose until intersect the 3D terrain model in the requested point. We tested the proposed approach on a synthetic and real data. Experimental results proved the feasibility and robustness of the proposed approach and the precision is the
same order as the 3D terrain engine.