High School Students Invent Thermal AI Dash Cam to Reduce Deer Collisions

A group of high school students from STEM School Highlands Ranch in Denver have invented a dash cam-style device that uses artificial intelligence to detect wildlife on the road and warn approaching drivers. The device uses an infrared camera to scan for animals in the road and a special AI model to identify them. Now in its final prototyping stages, the team of young engineers hopes the invention can be used by drivers everywhere to reduce the number of wildlife-vehicle collisions that occur each year.

The project began in September, when Siddhi Singh (then a sophomore) told computer science department chair Tylor Chacon that she wanted to apply to the 2024 program. Samsung solution for tomorrow competition, a nationwide challenge that encourages high school engineering clubs to come up with innovative solutions to real-world problems. The best ideas are rewarded with cash prizes, awards, and industry support.

Singh and her teammates Dhriti Sinha, RJ Ballheim and Bri Scollville decided that wildlife collisions were a problem worth addressing. About 5,000 wildlife collisions occur on roads in their state each year, costing about $80 million annually, according to the Colorado Department of Transportation reports. When we look at the country as a whole, the annual number of collisions with wildlife climbs to more than a million, with about 200 human deaths in a typical year, according to a 2008 study. That’s why deer are the deadliest animal in America.

“We all felt a connection to that issue because we live in Colorado,” Singh says Outdoor living“We have a lot of beautiful wildlife, but on the other hand we also have a lot of collisions. So it was a very important issue for us.”

Initially, the group wanted to create a sound device that would deter deer and other wildlife. But after talking to several experts, including CDOT and Colorado Parks and Wildlife, it became clear that deterrence would not be an effective way to prevent encounters with wildlife. (Have you ever honked at a deer and had it just stand there and stare at you?)

CDOT pointed out that their current method of avoiding wildlife-vehicle collisions involved placing lighted signs on roadways, using state funds at taxpayers’ expense. While these signs are useful for keeping drivers cautious in areas with high wildlife densities, they can’t be everywhere at once. The group wanted to create a device that would give drivers more ownership and individual responsibility in avoiding collisions, something that drivers could use wherever they went.

Read more: The Key to Resolving Big Game Migration Conflicts? Roadkill

“Before we finalized our design, we looked at current solutions,” Singh said. “It’s important when creating new technology to look at what’s already out there and how you can make it better. CDOT spent millions on these signs. They’re expensive, they’re stationary, and people sometimes just ignore them. So a personal device that alerts the driver right away would be much better than a stationary device that people would ignore.”

Every expert the group spoke with was quick to remind them of the challenges and obstacles to such an invention, Chacon says. For the device to function reliably in Colorado’s varying weather patterns, it couldn’t be triggered by rain, snow, sleet or wind-blown debris. It also had to work in the dark, when most wildlife-vehicle collisions occur. Infrared, or thermal imaging, used in many trail cameras and a growing number of optics (it’s also the subject of some debate among hunters), seemed like the best answer to all of these challenges.

“Infrared is a wavelength that many animals emit,” Scoville says. “It’s an efficient way to detect animals and other life forms, especially in low-light and dark situations. When you’re driving at night, you can’t see animals that you might hit. We want the device to detect wildlife [well] before the human eye could detect it, otherwise the device wouldn’t do much.”

The final design, which is still in the works, would use a Flir infrared camera to take a constant stream of images of the road ahead. The camera would feed those images through a special chip programmed with the AI ​​model Singh designed and taught to recognize deer shapes. The dashboard-mounted device would then emit a soft LED flash and sound to the driver whenever a deer appears in view. The group plans for the device to detect deer and other wildlife from about 500 feet away, giving a driver traveling at 80 mph more than 4 seconds to avoid a collision. While that may not seem like a lot of time, it would certainly give drivers an edge over the traditional method of avoiding deer collisions: wait for them to flash in front of your headlights, slam on the brakes, and hit them anyway.Traditional headlights (only extend to about 250 feet.)

The students’ latest prototype won the Samsung state competition. Their prize package included approximately $12,000 worth of educational technology, including laptops, a Smart Board, and a phone with a high-quality camera for their school. They distributed some of their prizes to other departments within the school and donated a cash gift card to a local elementary school to fund a school garden. Although they failed to win the national competition, this doesn’t bother the group much.

“We didn’t really lose,” Sinha says Outdoor living. “You lose when you give up on the project. But we didn’t give up. We’re still working on it and refining our idea, so we didn’t lose.”

A lower-end version of the device has already been tested on the school’s security dog ​​while mounted on a remote-controlled car. But the group has taken the device far beyond its initial design phase, Chacon says.

“We’re using cutting-edge technology and we’re in the final stages of prototyping,” he says. “We’re using a real chip that’s meant for AI models and a $300 Flir camera. It’s a real device, it’s not just the cheap infrared sensors and cameras that people saw on our first prototype. We have the horsepower to actually do this.”

Read more: Feds Plan to Spend $350 Million on Wildlife Crossings. Transportation Secretary Pete Buttigieg Tells Us Why

The group is currently preparing the final prototype and thinking about how to test it. They are also looking for more infrared trail camera footage of deer to further train the AI ​​model. The more infrared deer footage the group can give it, the more likely the model is to recognize a red spot with four legs and a head as a deer on the road. They hope to eventually train the model to recognize other species.

Chacon thinks that the responsibility of testing the device on a real vehicle near real wildlife might fall on him, since none of the girls have driver’s licenses yet. But he doesn’t mind.

“Seeing them use some of the skills they’ve learned in engineering and computer science to build something that no one has ever thought of before is the pinnacle of my work,” Chacon says. “This is the grand prize. I feel like I’ve won the championship belt of teaching.”