|
P. Eisert and B. Girod
4 Q- v% g- |3 w0 w- xUniversity of Erlangen-Nuremberg, Germany! t! ~% u/ y8 k' }. Y. o
$ V' V* A" T% J( S& oIn this paper we present a model-based algorithm for
0 \/ }8 {% h, ]4 E! Tthe estimation of three-dimensional motion parameters
: ~0 ~8 H7 l& i+ x8 Vof an object moving in 3D-space. Photometric
* H5 E: k, _* \3 n" E2 Meffects are taken into account by adding different illumination
& X6 u/ L! R1 G5 fmodels to the virtual scene. Using the additional6 E9 i: Q1 W7 `
information from three-dimensional geometric% f5 _: M2 t* S' }+ e
models of the scene leads to linear algorithms for. z4 l5 o% e' }
the parameter estimation of the illumination models3 ^" K' T! x- Z, ]6 V+ p7 I4 @
which are all computationally efficient. Experiments$ p o8 m. n, C) n
show that the Peak Signal Noise Ratio (PSNR) between
$ K2 d. a4 @7 Tcamera and reconstructed synthetic images can
, F- F6 p- x* fbe increased by up to 7 dB compared to global illumination
) f Y+ p, F- \% jcompensation. The average estimation error: d+ [9 D& `3 c: Y& j5 A7 M
of the motion parameters is at the same time reduced
9 C) `3 m3 n, ^+ Fby 40 %.. M% @0 L& G4 `$ D+ D% r! ]* f1 q
) R' e" j7 D7 x4 n1 m7 ~[ 本帖最後由 masonchung 於 2008-4-15 07:42 PM 編輯 ] |
本帖子中包含更多資源
您需要 登錄 才可以下載或查看,沒有帳號?申請會員
x
|