Counting turtles ...

[skyfilms]

Hi. Would anyone recommend a software or technique to count turtles from an aerial picture? The idea is to take pictures (from a P4) of a 1 square kilometer area (ocean, clear water, turtles can be seen easily from 40 meters high as higher contrast "ovals") and stitch the pictures, then scan for turtles, then repeat the next day and so on to get a good count of their population and migration trends in that South Pacific area.
Turtles are cold blooded (unfortunately thermal picture would not help) and are 50 cm to 2 meters is size.
Any suggestions please, for the image capture, stitching then the processing?
Thanks much
Patrick

 

[Meta4]

Hi. Would anyone recommend a software or technique to count turtles from an aerial picture? The idea is to take pictures (from a P4) of a 1 square kilometer area (ocean, clear water, turtles can be seen easily from 40 meters high as higher contrast "ovals") and stitch the pictures, then scan for turtles, then repeat the next day and so on to get a good count of their population and migration trends in that South Pacific area.
Turtles are cold blooded (unfortunately thermal picture would not help) and are 50 cm to 2 meters is size.
Any suggestions please, for the image capture, stitching then the processing?

This sounds like a tricky project.
Have you tried stitching ocean photos already?
Stitching requires features for the software to identify and match and that's going to be a problem with water.

You'll also have to watch the sun angle or you can just catch a lot of glare.

The turtle's visibility will vary with their depth and only turtles close to the surface are going to be easy to spot.
If there is enough contrast in a photo between the appearance of a turtle and the water around it, you might be able to find them with the magic wand tool in Phantoshop.
Untick Contiguous and experiment with Tolerance.

 

[skyfilms]

Hi. Thanks for taking the time to analyze this. I've never stitched aerial ocean picture manually, I previously used Pix4D and DroneDeploy but on land. I think these work with coordinates rather than finding edges to stitch, but I may be mistaken (and optimistic!)

Anything above the current method would be great (infrequent and costly airplane flyovers with the pilot counting as s/he flies...)

 

[Richard R]

Most stitching software warns against trying to stitch water. Lack of features and a lot of instability in the images. You may not be able to stitch stills, but you could run a Litchi (for example) mission over the area of interest and film video. You could film as much as 100 acres on a single battery. someone would have to manually watch the video and count animals. Wouldn't have exact geolocation though. Another approach would be to collect the stills (using something like Maps Made Easy app), then instead of uploading them to stitch, just again manually scan the pictures. Now, however, you could have the geo-coordinates of each picture to locate the turtles. As an example, I recently ran a 50 acre shoot using MME from about 400 ft AGL that took 100 pictures and had a GSD or ground level resolution of about 2 cm (1.9 inches)

 

[ecolumbia999]

Hi Patrick,

I might have a solution for some custom software. Coincidentally I'm working on a project to count turtle tracks and place points on a maps as the drone flies along. This is using computer vision to interpret the video in real time. This solution could also work from what you describe. Feel free to contact me at davejacob at silverdynsoftware dot com