It is certainly a thrilling time to be working in artificial intelligence, and this year at re:Invent was no exception!
Amid all the delight we found while discussing how Digital Nebula can partner to enable artificial intelligence applications within your business, AWS came out with the release of DeepLens, an artificial intelligence equipped camera! This camera has onboard connectivity hardware including a micro HDMI, two USB ports, and dual-band WiFi which enables internet of things connectivity. It carries computing power using the Intel Atom processor with 8GB of RAM, and the camera can shoot 1080p and has 4GB. But the most attractive artificial intelligence aspect of the DeepLens camera is the ability to use DeepLens to carry pre-trained intelligence models to use to explore artificial intelligence applications right within your grasp at a small price of only $249. As part of the release to get you started, Amazon has some already trained intelligence model templates that can be loaded right away such as facial recognition.
After the announcement, we have been excited to explore incorporating DeepLens into some of the solutions we have already built. First and most obviously, is the application of using DeepLens alongside our solution for facial recognition when the face is compared to a user’s license. The current solution (found here in more detail), asks the user to upload a “selfie” photo and to separately upload a license photo. When applying the features inherent in DeepLens, there is the possibility of using two cameras alongside each other: the first one pre-loaded with the facial recognition feature, and the second pre-loaded with license recognition modeling. Using those recognize model capabilities, DeepLens could capture the necessary photo automatically and then send it to the back end system for the comparison model and confidence analysis of the match.
Another possible solution along the same lines could be used in place of that TSA agent which checks your boarding pass name against your photo ID. Adding a third DeepLens camera pretrained to read the boarding pass name could also be checked on the back end against the other two cameras to verify the match to the license and the match of the license photo to the person standing there. Or maybe in your office you have someone checking badge names against licenses before a guest comes into the office. The DeepLens cameras could be pre-trained to read and register badge names and license names at the front desk and if the comparison process is successful on the back end, open the door for them automatically. Using a process like this would also keep photographic, timestamped logs of data for you, as an added security bonus.
Applying DeepLens alongside our drone solution, we could load our pre-trained intelligence model which recognizes the damage on a wind turbine or oil pipeline to identify that damage in real time while still up in the air, and take a desired action such as shooting extra close up photos of the damage identified.
For an inventory application, using a DeepLens pre-trained model to identify missing inventory could trigger the camera to store the UPC number and once connected back to the cloud at the end of the inventory process, could automatically feed those pertinent numbers to a system that orders them right away.
The power of DeepLens is in its ability to enable some pre-trained artificial intelligence to happen at the edge, limiting the travel time necessary for the device to call back to the cloud just to perform the initial computation for the intelligence model. It is the start of enabling small bots (DeepLens) with eyes (the camera) to carry intelligence to recognize and maybe even perform small actions right away in real time.
So what do you think? Do you have more ideas for how to test out DeepLens capabilities within your business? Connect with us to brainstorm ideas or to get started. We can’t wait to get started exploring further applications at the edge right away!