Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Messages - UAV_project

Pages: 1 [2]
After another day of working on this band alignment task, I am wondering if there is any way that I could ensure more consistent results from the camera optimization parameters?

In the end I would like to have it automated such that I don't need to go checking the result manually each time in order to see if the result was accurate or not. Because right now I am running the exact same code over and over, but sometimes it will align perfectly, while other times the results are a somewhat offset.

Well I have been having pretty good results now by using the coordinates of the point cloud tie points and feeding them into cv2 to find the homography and then warping. The only issue is that when I do get a good resulting homography matrix, I can't seem to obtain the same results when I repeat the process. I guess this is due to some randomness in the camera parameter optimization?

Hey Alexey,

The second option is what I am looking for. Essentially, what I have is a mask image which I have created to overlap with the MicaSense band 4. But since the other bands are not aligned with band 4, the mask is inaccurate for the other bands. So I want to transform the other MicaSense bands for this capture so that my mask lines up with all bands instead of having the make a new mask for each other MicaSense band.

Yesterday I was trying to get a set of tiepoints by loading just the 5 band images into Metashape and deriving the point cloud in hopes that the coordinates of the tiepoint projections could help me to get a homography matrix using something like OpenCV, but if there is a way to perform this alignment task using only Metashape then I would prefer this method.


Hey guys,

I am using Metashape for orthomosaics, but I was wondering if it could perhaps be used for another task in my project.
I have a set of MicaSense images where the bands are unaligned. What I want is to align all bands for a given image. I have tried to play around with it myself, using a 3x3 subset of the[0].transform as a homography matrix, but it doesn't seem to be working.

Is it possible for Metashape to help me with this task? Or shall I look elsewhere? I have also tried CV2 but they seem to have poor support for 16bit imagery (which MicaSense is), and my warped images are being given back as 8bit.


Python and Java API / Re: Quality measure of keypoints/tiepoints
« on: April 16, 2020, 12:20:08 PM »
Sorry to keep bumping this, but maybe it is not clear what I meant with 'Quality'. I know there are measures (reprojection error, reconstruction uncertainty, projection accuracy) that are associated to the tie points, but these are not what I mean.
I am wondering if there is also a 'dissimilarity' score which is based on the features of the pixels (shape, color, ect). This would help us to correct our images which were accidentally taken using automatic exposure settings (so exposures were not constant).


Python and Java API / Re: Quality measure of keypoints/tiepoints
« on: April 14, 2020, 09:41:11 AM »
Any ideas on this? From searching to forums it seems no one has ever asked this question before, and there is no mention within the user's manual, but maybe someone has a hack for how I could achieve this?

Python and Java API / Re: Gradual Selection Python Script
« on: April 14, 2020, 09:38:36 AM »
The problem is that you need to change

Code: [Select]
for i in range(list_values):

Code: [Select]
for i in range(len(list_values)):
in line 21 of your script

Python and Java API / Quality measure of keypoints/tiepoints
« on: April 09, 2020, 03:57:43 PM »
Hey guys,

I found in previous posts that I can use the projections of the tie points to find the pixels which match within different images. However, I am now wondering, is it possible for me to get some information on how confident the algorithm is that these points are matched? Something like a similarity score for a given projection track?


General / Re: Calculation of pixel size by Metashape
« on: March 16, 2020, 01:21:30 PM »
Any ideas? I am mainly just wondering if I should set these values with my own simple calculations or just leave it to Metashape? Since I just found this simple formula through googling I am not sure if it is the best method of calculation for pixel sizes.

General / Calculation of pixel size by Metashape
« on: March 13, 2020, 12:21:05 PM »
I have noticed that Metashape will calculate the pixel_width, pixel_height, and pixel_size seemingly on its own (using the python API) when no values are explicitly provided. The values are very close to what I have calculated myself, but there is still some discrepency:

My calculation :  0.003916667
Metashape calc : 0.00403846

So I am wondering, how does Metashape come up with this value? For my own calculation I used the sensor width or height in mm divided by sensor width or height in pixels.

Also, what would the effect be, using one over the other? Does the use of values which differ by such a small margin have any real effect on the output (DEM, orthomosaic)?


Pages: 1 [2]