Motion From Point Matches Using Affine Epipolar Geometry

摘要: Algorithms to perform point-based motion estimation under orthographic and scaled projection abound in the literature. A key limitation of many existing algorithms is that they rely on selection a minimal point set define “local coordinate frame”. This approach extremely sensitive errors noise, forfeits advantages using full data set. Furthermore, attention seldom paid statistical performance algorithms. We present new framework caters for allows all available features be used, without need select frame explicitly. theory derived context affine camera, which generalises orthographic, para-perspective models. epipolar geometry two such cameras, giving fundamental matrix this case discussing its noise resistant computation. The two-view rigid parameters (the scale factor between views, 3D axis rotation cyclotorsion angle) are then determined directly from geometry. Optimal estimates obtained over time by means linear Kalman filter, results presented real data.