Which alignment model should work for the Plant Health Camera

In remote sensing we try to keep the spectral information as pure as possible.

In some cases, that means that band data has to be adjusted spatially to align the spectral information properly.

Have you worked out the spatial accuracy of your camera alignment?

I suggest that you keep any alignment issues to a minimum.  You might find it better to take lower resolution summations using different spatial patterns to identify spectral issues.

There are many simple techniques to use, but you need to understand how they best represent your spatial and spectral information content.

DAB

Hi DAB

it is indeed important to get as pure as possible spectral information. And since I'm using different cameras for RED and NIR, the bands have to be aligned properly. Details here: ([Pi IoT] Plant Health Camera #8 - Aligning the images )

For remote sensing with a large distance from the sensor to the ground I expect translation would be sufficient, since all pixels are more or less at the same distance to the sensor.

For close proximity as in my application this is different.

For images inside, from 50cm to the plant tomography was needed in order to properly align pixels close and at larger distances to the cameras. For the outside image as can be seen in the examples in the blogpost translation seems to work fine.

You are right, I did'n quantify the spatial accuracy yet. I'm planning to do so, but don't like to update my blog posts as long as it is under evaluation by the jury of the challenge.

best,

Gerrit.

Hi DAB

it is indeed important to get as pure as possible spectral information. And since I'm using different cameras for RED and NIR, the bands have to be aligned properly. Details here: ([Pi IoT] Plant Health Camera #8 - Aligning the images )

For remote sensing with a large distance from the sensor to the ground I expect translation would be sufficient, since all pixels are more or less at the same distance to the sensor.

For close proximity as in my application this is different.

For images inside, from 50cm to the plant tomography was needed in order to properly align pixels close and at larger distances to the cameras. For the outside image as can be seen in the examples in the blogpost translation seems to work fine.

You are right, I did'n quantify the spatial accuracy yet. I'm planning to do so, but don't like to update my blog posts as long as it is under evaluation by the jury of the challenge.

best,

Gerrit.