Long Image Sequence Motion Analysis Using Polynomial Motion Models

The cameracentered motion representation is ([1][5][6]): This paper presents two algorithms for estimating motion and structure from long monocular image sequences: one using interframe correspondences, the other using point trajectories. Object-centered motion represenfations and motion models &scribed by up to the second order polynomials are used to estimate motion parameters from long image sequences. The algorithm automatically fvld the proper model that applies lo an image sequence and give the globally optimal solution for the motion parameters under the chosen model. The selection of motion model for rotation is independent from that for translation. Since the algorithms first solve for rotation parameters nonlinearly and then solve for translation and structure parameters in closed forms, they invlove less unknowns in the nonlinear search and are hence more robust and ejicient than existing long sequence algorithms. Experimental results with real image data are presented.