Direct estimation of 3‐d motion parameters from image sequence and depth

Masanobu Yamamoto · Systems and Computers in Japan · 1986

Abstract The methods to estimate the three‐dimensional motion parameters of a moving object from image sequence can roughly be divided into the two‐stage estimation based on the correspondence between the successive images and the direct estimation based on the time‐difference images. By applying the direct estimation method to the 3‐D motion, not only the motion information but also the 3‐D geometrical information can theoretically be acquired. It is difficult, however, to obtain a solution with a high accuracy, due to the nonlinearity of the computation and the assumption of small variation between successive images. The simplicity of the algorithm, which is the merit of the method, is also lost. From such a viewpoint, this paper first derives the linear equation and its solution for the direct estimation of the 3‐D motion parameters from the image sequence, by utilizing the depth information which is given by some range finder. Then the condition for the uniqueness of the solution of the equation is discussed. The necessary and sufficient condition is derived in geometrical terms for the image pattern to satisfy, so that the motion can be interpreted uniquely. The proposed method is applied to the actual image sequence, verifying that the 3‐D motion parameters can be estimated with a high accuracy.

Read the paper · More papers on PaperTik