Joint direct estimation of 3D geometry and 3D motion using spatio temporal gradients

Conventional image-motion based methods for structure from motion first compute optical flow, then solve the 3D parameters on epipolar constraint, and finally recover geometry of scene. However, errors in flow due to regularization can lead large structure. This paper investigates whether performance consistency be improved by avoiding estimation early stages structure-from-motion pipeline, it proposes a new direct method image gradients (normal flow) only. Our main idea lies reformulation positive-depth constraint – basis estimating egomotion normal as continuous piecewise differentiable function, which allows use well-known minimization techniques motion. The estimate is refined estimated adding depth. Experimental comparisons standard synthetic datasets real-world driving benchmark dataset Kitti using three different optic algorithms show that achieves better accuracy all but one case. Furthermore, outperforms existing techniques. Finally, recovered shown also very accurate.