search

hku-mars / BALM

An efficient and consistent bundle adjustment for lidar mapping
GNU General Public License v2.0
700 stars 151 forks source link

Why Second order approximation #9

Closed dysenustc closed 3 years ago

dysenustc commented 3 years ago

Hi, i have a question about the slide window ba, why you choose second order approximation eigenvalues w.r.t poses? I try to use first order approximation to solve this problem, but the iteration result cannot converge. Is this ba formulation converge region is small?

Zale-Liu commented 3 years ago

The cost function for second-order Newton iteration method and Gauss-Newton iteration method are different. The eigen value is equal to the square of distance. If you use Gauss-Newton method, it mean you optimize the fourth power of distance, that's why the convergence is very slow. But if you use second-order Newton method, the optimization variable is the square of disstance. It generally needs 4 or 5 iterations to converge.

dysenustc commented 3 years ago

ok, thank you very much