We present a translation estimation method from omnidirectional images with missing data. Given omnidirectional images, it is possible to estimate the rotations of each camera using a fundamental matrix. We can thus get the translations of each camera using a plane-based projective reconstruction method. However, the plane-based translation estimation method does not give a robust result when there are measurement errors in the data. The method proposed in this paper solves this problem and shows more reliable results than the previous method.