Research: A rigorous, realistic and reproducible test of camera trap perfomance

I am working towards having a test of camera trap performance where the targets are real, moving animals instead of people, the movement of the targets is controlled so that tests can be repeated and reproduced, and the movement of the targets is known so that missed detections as well as captured images are part of the data.

The aim is to detect deficiencies in camera performance that might make them unfit for purpose, and to rigorously compare camera makes and models.

I have been using a large, trained domestic dog as a surrogate for African wild dogs, leopards, cheetahs and other wild targets of similar size.  He walks a set pattern of tracks in front of the camera(s), by doing recalls across the field of view (FOV) (see the arena layout diagram in the attachments; each arrow is walked in both directions). I am comparing 4 cameras; an old Bushnell Trophycam, a new Reconyx Ultrafire XR6, a new Browning Strikeforce and a new Spypoint Force 11D

Pilot trials show that:

- the cross walk at 2m and the parts of the radial walks within 2m are uninformative- all the cameras capture images every time.

- duplicate tests on the same day show good repeatability both overall and for each walk, so the dog is behaving consistently enough for the tests to be reliable.

- tests on different days with exactly the same setup consistently assigned the same performance ranks to the four cameras under test for each aspect of the test. Nonetheless a small change to the angle at which the Spypoint was mounted moved it from the bottom to the top of the rankings for range and number of images. This confirms earlier findings that image capture rate on both Spypoint and Reconyx is very sensitive to small changes in camera height and angle. I plan to test this for the Browning and Bushnell, and test the stability of the new rankings. Different cameras show clearly different strengths and weaknesses; the Browning has less range than the others, but captures more images than the Reconyx and Bushnell, and is more consistent than the Spypoint. The Reconyx captures fewer images than any of the others.

- with a few exceptions the crossways walks (A, B, C, D, E) show symmetrical responses; the number of images captured is the same for walks from right to left as from left to right.

- all four cameras have very low detection rates for radial walks (arrows H and I), and a strong bias for there to be 2-3 times as many images captured when the dog walks away from the camera than when he walks towards it.

- there appear to be gaps and hotspots in the detections by all four cameras.

- once the ambient temperature rises to 30 C all the cameras become insensitive.

This test shows promise as a way of rigorously comparing cameras. The detections being biased by direction of movement, and the presence of dead zones and hot spots may violate assumptions about detectability that underpin the use of camera data for estimates of animal distribution and abundance.

wlphipps's picture
StephODonnell's picture
Peter Apps branched this thread at this point into Update on camera trap tests with real animal targets