Issue link: https://iconnect007.uberflip.com/i/1172746
28 SMT007 MAGAZINE I OCTOBER 2019 age accompanies each image sent to headquar- ters as well as a health check, including bat- tery voltage. Our mantra in designing and building this device to work in Africa might resonate with some of your designers; our five-part mantra, in order to scale, is it has to be cheap, dura- ble, easy-to-use, efficient, and if it involves a battery, low-power. If it doesn't meet all five of those criteria, forget about it; it won't be used nor be scalable. We've had to check all of those boxes and engineer our technology to optimize on all of those fronts. The Myriad 2 chip helps us by being low-cost and low-power and has a very small form factor. The other crucial piece about these cameras is that they're very small and easy to hide. Most wildlife cameras are bulky and conspicuous; some studies have shown that in the first year of operation for these types of cameras, 42% were either vandalized or stolen. That doesn't do you any good if you're spending a lot of money on these cameras and they have a short lifespan. Ours have yet to be found because they're so easy to hide. The camera and com- puter together are the size of your index fin- ger. Also, the device is linked to a multipur- pose communications unit that's three inches by one inch. It's so tiny in comparison to other cameras sold commercially. Then, there has to be connectivity. We've solved the battery life issue by using inference on the edge and filtering there rather than in the cloud. And because we don't have to send all of those images to the cloud, the most expensive thing is the transmission of the images, so we save on that. But at the end of the day, if that's all we did, we'd have a smart camera that could run a bunch of different deep, neural network models, but it wouldn't be a burglar alarm. You'd still have to go to the site to remove the SD card, down- load the images, and then see that a poacher came through five days ago, for example; instead, you want to be out in near real-time and not just be aware of what already happened. As mentioned earlier, we've linked our TrailGuard AI camera to a communications unit that has Wi-Fi and a GSM and LoRa modem that looks for GSM first, which has a range of 35 kilometers. If there is a GSM range, it will send the image in under two minutes to head- quarters. If there's no GSM, which is typical for a lot of parks around the world where the cell coverage stops at the edge of the park, we use a radio link. As long as the image is below 20 kilobytes, we can send it over the radio link with up to 30 kilometers line of sight, or about 10 kilometers not line of sight. Then, it goes over a LoRa modem to our "WildTech gate- way," which allows many TrailGuard AI units to communicate with a single satellite modem. As Intel has helped us on the AI side in power consumption issues, Inmarsat, the leading sat- ellite company, made a philanthropic contribu- tion to help us solve connectivity issues. Johnson: What does that look like, Eric? Dinerstein: We use a state-of-the-art, military- grade satellite modem—the BGAN Explorer 540—that connects with our little gateway, which is two inches by one inch by one inch. That allows multiple TrailGuard devices to send images to a single satellite modem and amortize the cost of the satellite modem over many devices. Galaxy 1, through Inmarsat, has also offered us a reasonable transmission data rate that makes this affordable. After the ini- tial capital cost of the satellite modems and the TrailGuard AI units in a park, we estimate