 This week's IonMPI brought to you by DigiKey is Intel's RealSense. That's right. Check out that cool new Intel logo. So this is the RealSense family. We actually covered RealSense LiDAR cameras like eight, nine months ago, but they came with a new product and I thought it was cool. Now, people who watch this show know that I am a total sucker for sensors and edge machine learning devices because I really like machine learning, but I also think privacy is very important. So this is an identification system, an official identification system that's completely on the edge and very affordable. So this is the module. This is the F450, not to be confused with the truck, and the F455, which is the enclosed version. Same technology, but one of them is the bare PCB and one of them is enclosed. So I played with the enclosed one because it's, of course, easier to use, but for your final product, you'd use the bare PCB. Even the enclosed one, it's got a lot of mounting holes. Look at how nice this is, a full aluminum enclosure that's easy to use. Come for the kit with a little tripod, USB cable, starting pack. So what is this? This is a depth camera. It's not a true LiDAR. It uses, I think, an infrared scatter and two cameras to do depth recognition, to do facial authentication on the edge. So you don't need to be connected to the internet. You don't need to have a cloud computing thing. Everything is stored on the device and it's really easy to use. And what I thought was really cool is they specifically talk about, I mean, some people might say this is like, it's accessibility, but it's still like recognizing faces and they don't like any kind of facial recognition. But that said, it does work on a wide range of people with different hairstyles and facial hair and glasses or prostheses and works for people of like a variety of heights. So if you're going to do a facial recognition system, I think this is the least bad, right? Because there's no cloud and it's tested to work with a wide range of human faces, which is good if you're doing that sort of thing. So this is what's inside of it. So there is a chip, the flash memory, the projector, that's probably the infrared projector, the illuminator, IR illuminator, IR LED. And two cameras that you do depth that lets you do basic depth mapping and avoids people like holding up a photo of your face and having that work. There's also a secure element. This actually allows you to, you can't communicate with it over USB or you are using encrypted data packets. I won't cover that here, but it's in the SDK and the datasheet. So if you're like, I want to have everything fully encrypted, you can do that. So even if you're sniffing the connection, you wouldn't be able to see what's going on. What's neat, I think, about this is on the board itself, you have two ways of connecting. You can power it from three to five volts. This means it can be powered from USB or from a LiPo battery, a couple of nickel metal hydride batteries that are rechargeable. There's UART and USB connectivity. The USB, of course, gets you full video. But if you just want to control it on a device, like as an embedded device, you can use UART. So you can use it with a microcontroller or a microcomputer. That's what it looks like. There's a full datasheet with all the connectivity. Bottom right is that flex connector, which has like 54 pins. And all of the pins will let you basically reach in and basically get the video or get the infrared data or get like the point map from the infrared. Here you can see the projector and the LED are 850 nanometer IR, the IPEX host connector for, again, embedding it. Or if you want to use it with a computer, like if your system is using a normal desktop computer or a single board computer, just use USB. This was the FPC connector. And then all the pins. So you can look. I'm not going to go through all them. Everything is brought out. So you can integrate this very deeply into your product using a flux connector. So all that hard stuff to take care of. Here is some examples of, it says audio, but it doesn't speak. The audio will be handled separately because it doesn't have a speaker on the module. But you can have it speak. You can have, here's some of the error codes you can get, how to communicate with the user, to enroll people without a computer, and also detect faces without a computer. There's a full SDK. Intel RealSense has a GitHub repo. They have SDK and example code for a variety of use cases and different languages, which I thought was really cool. One of the things that I really liked about this is it's like really embeddable. It's all done for you. And the SDK is simple but effective. It's just all the things you want, and none of the nonsense that you don't want to deal with, so you can really quickly integrate this into your product. This is a software architecture. I think in here, I think it covers some of the encryption stuff and all the stuff that you can touch. Again, I'm not going to talk about the encryption, but you can have it be fully encrypted. And it doesn't store photos of faces. It stores only points, vectors of how it identifies the face. It also does a fuzzy learning. And so if somebody's face ages, or they're wearing glasses, or they're not, or they grew a beard, or they're not, or their hair's up or down, or they're wearing a scarf, or wearing a mask, it can still identify faces. And we'll show that in a little video. You can upload. So download it. If you want to get started quickly, with Windows, there's an application. You download it because this is new. I updated the firmware. There's a console code, so you can actually see the UART data sent back and forth, which could be really good if you are like, I want to debug my UART code and how is the application working, so you can see all the packet data being sent. I enrolled my face. Basically, you just get it nice and centered, tilt to the left, tilt to the right. It did take me a couple tries to get used to how it wants you to scan, but I think a lot of times when you're enrolling people, there's somebody there to get you enrolled, like it's a security system. And so it's not so bad, because once you know how to do it, it's very easy to tell people how to do it. But I definitely think you should enroll with somebody around to start and get used to it when you're starting to enroll your group of people. As it's enrolling, you can see the video output, and you can see it's purple because it's doing infrared and point cloud stuff. And then once you do that, it can authenticate you, so you can just say, like, detect who it is. Boom, lady detected. And as I moved around, it kind of followed me with the green box. So it did a pretty good job of detecting. And detection is very fast. It's under a second. So. Available on DigGate. Right, you can get the module or, again, like I said, the all-in-one starter kit. You're not going to be able to get anything cheaper than this. It's under $90 for a full, ready-to-go, two cameras, plus IR projector. I mean, this is the cheapest edge machine-learning face recognition system I've ever seen, that you have to do no work to integrate. This one didn't make me feel bad. No IDE was installed. This cloud thing is going to be bad? No, there's no cloud thing. This is good. And what's nice is that because you can get the data over UART, the rest of your system, if it's a security system or a lock or authentication, whatever, it doesn't have to be that complicated because all the really heavy-duty computational stuff is outsourced to this module, you can run this with an Arduino. Like, this could be plugged into an Arduino if you wanted. Ape and microcontroller could communicate with this because it's all over UART. Or if you have a full computer plugged in over USB, mount it, and it's all weather-proofed and nice to go. So you have two options. Did a very good job here. Easy projects with this, too. Like, flip a relay to open up a box or something. So here's a video. Trivial to do. This is their presentation that they did at CES that was only a few months ago. This is the engineer showing, of course, there's the mask demo. It works even when you have a mask working. Nobody has to do a mask demo now. And this is interesting. They actually showed it at slow speeds. You can see at half speed what is going on. It illuminates the area with LED. It shows this guy. I don't know if he grew a beard or he pissed one on, a mustache, and glasses. It still works with headphones. I'm pretty sure that's a fake mustache. You don't think he grew it in like five minutes? No. Oh, man. Different angles, one of the things that they really focused on was even in different light, different angle, it'll still work. One of the cool things about Edge is you don't have to worry about if your internet goes down, can you not get into your apartment. But this worked really well showing that it can't be spoofed with a cell phone image, because that's some old facial recognition systems. You could do that. Pieces of paper or cell phones won't work. It's going to illuminate it. It's going to be like, no, there's a flat object in front of me. It has to be a human-shaped head. OK, so that's OK. And that is this week's IonMPI. OK. IonMPI.