accuracy of 3d model in z direction??

[ahmed KIMOO]

hello,
i'm use agisoft photoscan for uav , after using control points the accuracy of check points in x and y were good but in z was not accurate ( error = 20 cm )
i tried for many project and thr result was the same
what is the problem for this??
how can reduce the error in z direction??
thanks,

[ahmed KIMOO]

?

[JMR]

can you share your flight design parameters and camera setup settings?, error in Z should be on par or just a bit worse than XY errors. it is typically just about 1.5 times higher.

[ahmed KIMOO]

can you share your flight design parameters and camera setup settings?, error in Z should be on par or just a bit worse than XY errors. it is typically just about 1.5 times higher.

Processing Parameters
General
Cameras 70
A ligned  cameras 70
Markers 12
Coor dinat e  sy st em W G S   84  (E P S G : : 4326)
P o i n t   C l o u d
P oint s 139, 310  of   170, 978
RM S   r epr ojec t ion  er r or 0. 207879  ( 1. 1 1589  pix )
M ax   r epr ojec t ion  er r or 1. 161 1 1  ( 40. 9082  pix )
M ean  k ey   point   siz e 4. 45752  pix
E f f ec t iv e  ov er lap 4. 19256
A l i g n men t   p aramet ers
A c c ur ac y High
P air   pr eselec t ion Disabled
K ey   point   limit 400, 000
T ie  point   limit 10, 000
Const r ain  f eat ur es  by   mask No
A dapt iv e  c amer a  model  f it t ing Y es
M at c hing  t ime 1  hour s  51  minut es
A lignment   t ime 4  minut es  36  sec onds
O p t i mi z at i o n   p aramet ers
P ar amet er s f ,   b1,   b2,   c x ,   c y ,   k 1- k 3,   p1,   p2
O pt imiz at ion  t ime 26  sec onds
S o f tware
V er sion 1. 2. 6  build  2834
P lat f orm W indows  64  bit


Nu mber  of   i mages: 70
Fl y i n g  al t i t u de: 175 m
Grou n d  resol u t i on : 5. 33  cm/ pi x
Cov erage  area: 0. 0743  km²
Camera  st at i on s: 70
T i e  poi n t s: 139, 310
Proj ect i on s: 536, 744
Reproj ect i on   error: 1. 12  pi x

[ahmed KIMOO]

hello,
i worked in some projects in agisoft photoscan
the result is good in x and y direction but in z the error was very large (30-40 cm)
what is the problem and how to reduce that.
thanks and sorry for annoying you.

[Alexey Pasumansky]

Hello ahmed KIMOO,

For aerial surveys the error along Z axis is usually higher than in the XY plane due to the vertical shooting scenario.

Errors can be reduced if control points are used for the project, in this case we usually suggest to uncheck all the cameras in the Reference pane prior to the optimization procedure. Of course, it is necessary to make sure that all the marker projections are placed  accurately and the marker accuracy is defined according to the measurement accuracy in the Reference pane settings.

[ahmed KIMOO]

Hello ahmed KIMOO,

For aerial surveys the error along Z axis is usually higher than in the XY plane due to the vertical shooting scenario.

Errors can be reduced if control points are used for the project, in this case we usually suggest to uncheck all the cameras in the Reference pane prior to the optimization procedure. Of course, it is necessary to make sure that all the marker projections are placed  accurately and the marker accuracy is defined according to the measurement accuracy in the Reference pane settings.

error (pix)= 0.002 in average
error in x= 0.02m
in y= 0.04 m
but in z= 0.35 m (very large )
i'm already do all you say above
but no good result
thanks,

[Alexey Pasumansky]

Hello ahmed KIMOO,

Can you share the report file generated from this project?

[ahmed KIMOO]

Hello ahmed KIMOO,

Can you share the report file generated from this project?

[Alexey Pasumansky]

Hello ahmed KIMOO,

The check points that you are using for the accuracy validation are outside the region covered by control points, also GCPs are not spread on the area of interest and are very close to each other.
Additionally the cameras are not unchecked in the Reference pane, so they are also used in the georeferencing slightly affecting the result due to the low accuracy of the on-board measurement device.

[ahmed KIMOO]

Hello ahmed KIMOO,

The check points that you are using for the accuracy validation are outside the region covered by control points, also GCPs are not spread on the area of interest and are very close to each other.
Additionally the cameras are not unchecked in the Reference pane, so they are also used in the georeferencing slightly affecting the result due to the low accuracy of the on-board measurement device.

can see it also

[Alexey Pasumansky]

Hello ahmed KIMOO,

Suggested processing settings for you set:
Align Photos: high accuracy, generic or reference preselection, 40 000 key point limit, 10 000 tie point limit, adaptive fitting - disabled.
Optimize Cameras: check on F, Cx, Cy, B1, B2, P1, P2, K1, K2, K3.

In your second case lowest alignment means that PhotoScan uses downsampled images (by factor 8 by each side). Using 400 000 for key point limit is unreasonable. While some parameters of the camera interior orientation remain unoptimized (the residuals remain high).

[ahmed KIMOO]

Hello ahmed KIMOO,

Suggested processing settings for you set:
Align Photos: high accuracy, generic or reference preselection, 40 000 key point limit, 10 000 tie point limit, adaptive fitting - disabled.
Optimize Cameras: check on F, Cx, Cy, B1, B2, P1, P2, K1, K2, K3.

In your second case lowest alignment means that PhotoScan uses downsampled images (by factor 8 by each side). Using 400 000 for key point limit is unreasonable. While some parameters of the camera interior orientation remain unoptimized (the residuals remain high).

thanks,
i will try with ur advice