search

nmoehrle / uavmvs

UAV capture planning for MVS reconstructions
Other
54 stars 19 forks source link

optimize_trajectory not update camera views correctly. Any suggestion? #16

Open szx0112 opened 3 years ago

szx0112 commented 3 years ago

Hi, I am able to generate the nadir_trajectory through 'generate-trajectory', then I use the nadir trajectory to compute the optimal traj using 'optimize_trajectory'. Figure below shows nadir and opt traj of NY-1 case

Nadir Traj (151 cameras) uav1

Opt Traj uav3

We see that the updated cameras are not update correctly. And the optimization step in around 2 seconds with return: 0 3 0 2.22444 1 2 0 2.22444 What is the meaning of the return?

I also tried the function plan_trajectory, but fails to get any camera view. The parameters used here are same as in the makeTraj.sh.

Any comment is appreciated.

laurelkeys commented 3 years ago

Hi @szx0112 were you able to get any output from plan_trajectory?

I ran generate-trajectory but, as you show here, it only generates a "grid-like / planar" trajectory. I was trying to find out which file generates the trajectories as shown in the paper.

Could you give us any directions @nmoehrle ?

nmoehrle commented 3 years ago

Using first 'generate-trajectory' and the 'optimize-trajectory' is how it is intended 'plan-trajectory' was merely an experiment. The output shows first the indes of the iteration, then how many cameras have been optimized and the overall volume of the simplexes (tetrahedra) that are used in the optimization. Since that is zero it terminates immediately. Did you adjust the distance parameters to your scene?

szx0112 commented 3 years ago

@laurelkeys Sorry for the late reply. Unfortunately, I haven't generate any optimized trajectory yet.

Thanks for the reply @nmoehrle . Yes, I tried and adjusted the parameters in makeTraj.sh. It seems that the parameters (ele/alt) need to be tuned for different synthetic scenes (UK-1, CA-1, ...). The nadir flights may too high/low for certain scenes.

I believe the distance parameters you mentioned (maxD, minD) defines the close/far plane of the camera frustum. I have tried the different combinations of distance parameters but found no luck (optimization always terminates in 2 iterations with zero simplexes return).

Do we have to tune the parameters for different scenes? And are there any recommended parameters for certain scene? Thanks.