H-023
Motion Estimation on Fish-Eye Images Using Modified Motion Model

○Arief Suryadi Satyawan（Waseda Univ.）・Junichi Hara（Waseda Univ./Ricoh）・Hiroshi Watanabe（Waseda Univ.）

Estimating motion directly from 2D fish-eye images can be done by applying differential optical flow concepts. However, due to massive distortion caused by the characteristic of fish-eye camera which results unique deformation of object's motion during captured by camera, the accuracy of vector flows yielded seems to be unsatisfying. To improve this, in this research, the motion model constraint is modified by incorporating a parametric motion model of fish-eye lens before finding the vector flows by adopting Lucas and Kanade's differential calculation. The results show that this approach can increase the performance of the reconstructed images around 1.5 dB.