So before I start this, please let me know if I am reiventing the wheel. If there are already reasonable solutions out there please stop me and let me know!

Our farms have cattle and sheep. I would like a drone similar to the LA100 series from Lehmann Aviation. Instead of checking out crops.

I would like the farmer to be able to open an easy to use app, click on the section of his farm he wants checked out and send it off.

The drone must be able to count livestock in camps ( a camp is a fenced off area. Farmers cycle their animals through camps to ensure that there is food and environment sustainability). 

I dont know if this would be easier/cheaper with IR or thermal imaging. I dont even know if you could put an IR reader that is sensitive enough on a drone so it doesnt have to fly too low so that it doesnt startle the animals or become a safety hazard.

Perhaps a thermal imagining drone would be better. If its possible to get accurate stock count of sheep and cattle from a camera. 

The bonus is the drone with a thermal camera can back up as a security drone to aid in anti-poaching activities at night. perhaps it can go up and use pattern recognition to try spot moving humans on your property. 


Which do you think would be better? thermal or IR? I will be using fixed wing (as farms here can be big)

Views: 3870

Replies to This Discussion

Your contributions are very informative :D

I want the end machine to use under $9000. Be durable and accurate. The multi role purpose will be important for farmers here. If it can count stock and keep a look out for farm attacks in anti poaching ops, then that would be great. Farming in SA is different to farming in the USA. We have constant attacks on farmers here by organized gangs. 

It would be a useful 2nd and similar function for it to be sent out every now and again when a farmer has heard reports of invaders on his farm. 

Another thing I want to know, who specializes in the software that counts animals? Is there a FOSS alternative yet? :)

@David, I can count your stock remotely. All I need is to set up server to run image processing and pattern recognition software. You would be able to access server remotely for image upload and count download. Thermal and IR imaginery is generally low-res so just take some shots with your standard camera and send to me. To run pattern recognition I need a number of samples to set correct resolution, size to limit processing time and get exact count number. Count should be verified by a human if landscape is not clear populated with static obstacles. I need res of your images to match ability to separate 2 closely positioned cows or sheeps. If you mix cattle and sheep at the same camp so downward pointing camera should shot images at altitude to let application to identify every objeect on the ground as an individual animal, removing shading problems. You can buy commercial drone at $9000 to work for you to accomplish this task.

Please email me

darius

manta103g@gmail.com

@David

Here are some of the ideas we built around.

ERDAS IMAGINE 2015 (Conservationdrones.org counts Palm trees from a single picture)

Palm Tree Example

We are working on a process using ERDAS IMAGINE 2015 to count cotton bolls per picture.

OpenCV items

Discussion 1

OpenCV detecting the hip bone of a cow passing by.

Scott Rapsey's OpenCV Cattle Detector Code

Study performed with actual results using a drone and thermal.

Generally speaking, OpenCV is not fit for counting people not to say about counting animals.

https://m.youtube.com/watch?v=WdfPN6e8CAo

Pattern object recognition apps work fine in case of flat images and fail to work

for 3D to 2D plane projected images.

Better approach is to process video imaginery to detect moving objects and calculate Z-dimension.

OpenCV fails to detect fuzzy objects by shape description not to mention fuzzy objects described by topology.

@David,  could you attach some images from your cattle farm ?

Thank you @Cala,

I played a lot with ImageJ, developing 3D USG Tomography.

Attention!!!
State of the art in computer counting of animals:

"
You could also use Amazon's Mechanical Turk.  You upload the pictures and pay someone like 15 cents to count the cows. You will get much more accurate numbers."

Hi - my company IDS applies deep learning to recognise features in images.

We recently took an orthomosaic of drone captured images of a sheep farm and used our neural network to train a classifier to predict the number of sheep in the image.

Rather than relying on tags, or thermal images, our system learns what a sheep looks like from the air ( height above ground ca. 100M ) and produces a count. see sample.

Currently images are uploaded from SD card to our cloud though we are looking at ways to do local processing.

Attachments:

RSS

© 2017   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service