Animal paparazzi: Camera traps focus on small critters

  • Cheaper tools and new techniques allow researchers to use camera traps on a broader array of animal species.

  • MotionMeerkat, an open-source computer vision program, can help scientists sort through abundant image data.

  • As always, researchers need to be sure they’re using the right tool to answer their particular question.

If you’ve ever tried to photograph a hummingbird, you might have noticed two things: those little birds are tiny, and they are fast.

Hummingbirds exemplify some of the challenges facing researchers trying to adapt camera trap technology to smaller targets than the deer, tigers and wildebeest usually found in their crosshairs. But their lower cost, greater sampling power and increased capacity and battery life make camera traps increasingly appealing to researchers studying small, rare and hard-to-access critters.

Too much of a good thing?

Ben Weinstein, a PhD student at Stony Brook University, studies the interactions between hummingbirds and plants in Ecuador, and he is deeply interested in how emerging technology can help scientists effectively study the environment.

This image of a hummingbird in Ecuador was captured using a camera trap. Image courtesy of  Ben Weinstein.
This image of a hummingbird in Ecuador was captured using a camera trap. Image courtesy of Ben Weinstein.

A traditional technique for studying hummingbird-plant interactions is to sit by a flower and record every bird that visits — or, if you’re well-funded, to pay someone else to do it for you. Apart from being mind-numbingly boring, the technique is expensive, fallible and increasingly difficult to achieve.

“The challenge we have is that there’s just not enough of us to be out in the field, and the incentives for academic research no longer align well with being out in the field all the time,” Weinstein said.

Given those constraints, “how do we sample in time and space sufficiently to actually observe biological phenomena? The only answer, really, is by embracing new technologies.”

So, Weinstein thought, why not set up a camera trap by the flowers and let a machine do the work? But the tiny birds would either fail to trigger the camera trap’s infrared sensor, or they’d flit through the frame too quickly to be captured in an image.

Then Weinstein tried an alternative strategy: He made a few modifications to a Plotwatcher Pro game camera to produce a device that snaps a photo every second from sunup to sundown and can capture three to four days’ worth of images. The camera worked well as a field-based sampling technique, but it created a new problem — thousands upon thousands of photos, only a tiny fraction of which captured hummingbirds.

The solution? MotionMeerkat.

Coded vision

Weinstein turned to computer vision and developed a program that could sort through the deluge of camera trap data and pull out images that were different. The algorithm used a technique called background subtraction, which, given a group of photos and a set threshold of difference, compares the images and flags pictures that cross the threshold.

The MotionMeerkat program analyzes the video stream and captures just the “events of interest” (e.g. images containing your target animal).

“We normally can throw out about 99 percent of the data,” Weinstein said, which could save busy researchers a lot of time sorting through photos. Also, because the camera is set to capture an image every second, the program can be analyzed using different levels of sensitivity. If an animal fails to trigger a normal camera trap, a researcher would never know.

This sample result shows just the clips (from a longer video) that actually contained sharks and other animals:

MotionMeerkat is available free on Weinstein’s website, and researchers have already used the program on studies involving sharks and desert gerbils.

Keeping an eye on rock rats

The critically endangered central rock-rat (Zyzomys pedunculatus) lives halfway around the world from Ecuador’s dazzling hummingbirds in the arid quartzite mountains of central Australia. But, like the hummingbirds, the rare rodent is also being studied by camera-trap-wielding researchers. Peter McDonald, of the University of Sydney and the Department of Land Resources Management, had been learning about the nocturnal and elusive rock-rat by capturing them in live traps, but the process was extremely labor- and resource-intensive: The rat’s remote and rugged habitat can often only be accessed by helicopter, and the researchers were unable to sample more than 10 sites at a time.

“Live trapping is inefficient, and we’d reached a point where our sampling method was hampering our ability to learn more about the species,” McDonald wrote in an email. “The massive advantage for us using cameras is that we can now run up to 100 sites concurrently, with a camera at each site.”

But the team faced a few challenges along the way. McDonald knew about researchers from Deakin University in Victoria who had experimented with camera traps for small- and medium-sized animals. Following their lead, McDonald planned to mount the cameras with their sensors pointing toward the ground instead of positioning them to capture images horizontally. But his team needed small, lightweight and sturdy posts that could be transported efficiently in a helicopter. They eventually settled on aluminum fence spacers, called droppers or stays. Although the minimal focal distance on some of the cameras they tested was too long when the cameras were mounted on the droppers, the team found commercially available cameras that suited their needs.

A camera trap in the unforgiving terrain of the MacDonnell Ranges, Australia. Image via McDonald, Peter J., et al.
A camera trap in the unforgiving terrain of the MacDonnell Ranges, Australia. Image via McDonald, Peter J., et al.

A few cameras failed in the field, but the project has been a success overall, providing detailed data on multiple rock-rat populations. The new technique “allows us to apply robust statistical analysis to our data,” McDonald wrote, “including occupancy modelling, which forms the basis of our monitoring program and efforts to learn more about the factors driving rock-rat occurrence.”

To camera trap, or not to camera trap?

There are a few considerations researchers should bear in mind when considering camera traps to photograph small-bodied species, according to McDonald and Weinstein.

Camera traps are not appropriate tools to determine species occupancy in all situations: The device must fit the scientist’s research question. If the subject is extremely rare or if it is found over a wide area and does not frequent a central spot, then camera traps might not be the best option.

And although it seems obvious, “Your organism must be identifiable by images,” Weinstein said. He’d thought that entomologists would be clamoring to use MotionMeerkat, but many species of insect pollinators are difficult to distinguish by images alone.

This image of a hummingbird in Ecuador was captured using a camera trap. Image courtesy of Ben
This image of a hummingbird in Ecuador was captured using a camera trap. Image courtesy of Ben Weinstein

The same goes for mammals, McDonald wrote: “The main potential limitation with using cameras for small mammals occurs when there are morphologically similar species. Sometimes using a white flash may help, but there are situations [where] camera traps will not allow observers to reliably identify species.”

Perhaps most importantly, researchers should realize that they often don’t have to reinvent the wheel. Computer vision has advanced immeasurably over the past few years, and researchers and conservationists should look for collaborators instead of trying to duplicate efforts.

“It’s not like we have to build anything ourselves,” Weinstein said, “It’s already done.”



McDonald, Peter J., et al. “Landscape-scale factors determine occupancy of the critically endangered central rock-rat in arid Australia: The utility of camera trapping.” Biological Conservation 191 (2015): 93-100.

Weinstein, Ben G. “MotionMeerkat: integrating motion video detection and ecological monitoring.” Methods in Ecology and Evolution 6.3 (2015): 357-362.

Article published by