How much camera movement is too much?

[NATHAN THAVARAJAH]

Hi Folks, I am just in the process to uploading my images to Agouti for REM analysis. I am noticing that many of our camera deployments have very slight movements of the cameras, between deploying and collection, it seems to be due to settling of the cameras over time. I am just wondering if even the smallest amount of movement means that the first calibrations are un-usable for animal detected before the movement. It isnt very clear on the training video how much movement is acceptable before precision of the calibrations are compromised. I attach an example. If you look at the trees on either side of the image you can see how much the camera has moved. Thanks, Nate 

[Marcus Rowcliffe]

Hi Nathan

I'm afraid there isn't an absolute cut-off for deciding how much movement is too much, but the rule of thumb is no more than a few pixels, particularly in the y (vertical) axis. In your example there is a shift of about 16 x pixels and 10 y pixels, which I find just about acceptable but wouldn't want to accept much more.

I hope this helps

Best

Marcus

[NATHAN THAVARAJAH]

Hi Marcus, thanks very much for your reply and advice. Please could i just ask one follow up question. With the example that I sent, although you would find it border line acceptable, if we have good calibrations at the beginning and end of deployment, would you prefer to spend the time to find when the camera moved and separate the detections accordingly for digitisation?

[Benjamin Evans]

Hi Nathan,

In this example, the slight change I don't think would make much difference. The biggest problem tends to arise from a change in the camera's pitch (up/down), as this shifts the horizon line and the distance from the camera forms an exponential curve, leading to a large change in distance estimation, particularly for animals further from the camera.

When we've got cameras that've been knocked more severely than in your example, we tend to annotate the side of the knock that contains the most observations, where there's reasonable calibration, and annotate the rest without tracking points. Another option, if you have multiple calibrations (e.g., deployment and collection), is to split them into multiple deployments at the same location, on either side of the displacement.

We've (Institute of Zoology, ZSL) been working on methods to automatically correct for minor knocks. We haven't finalised a process to share right now, but if you'd be happy to contribute your data, we'd greatly appreciate it! - Benjami...@ioz.ac.uk.

Hope that's somewhat helpful, though, in the meantime.

Best wishes,

Ben

https://bencevans.io/research

On Monday, June 1, 2026 at 8:00:03 PM UTC+1 NATHAN THAVARAJAH wrote: