The Performance of Camera Translation Direction Estimators From Optical Flow: Analysis, Comparison, and Theoretical Limits

A noniterative method using optical flow to recover the translation direction of a moving camera has been previously proposed by Heeger-Jepson (1992). In this paper we present a detailed explanation of the bias in this algorithm and compare methods for eliminating this bias, as well as presenting a comprehensive error analysis. This analysis includes a necessary modification to the Cramer-Rao lower bound (CRLB), We propose a simple iterative modification to the algorithm which produces unbiased translation direction estimates that approach the CRLB. Numerical results are used to compare the various techniques on synthetic and real image sequences.