字幕表 動画を再生する 英語字幕をプリント All right hey it's me Destin welcome back to smarter every day guns are a very divisive topic today's video is going to consider the problem of gun violence from an engineer's perspective the fact that I'm about to talk about guns online feels like I'm about to walk right onto the top of a huge active hornet's nest but I personally feel this conversation can take place in an intelligent respectful way I would love your help distributing this video because just by the very nature of the topic I've got hundreds of images of guns in this video it would not surprise me in the least if it was suppressed in the YouTube algorithm because YouTube itself doesn't want gun violence out there so please consider sharing this video if you like the discussion that we have here in a second no matter what your position is in the gun debate whether you love guns or you don't like guns I'm personally a gun owner we all agree and everyone I've spoken to agrees that we want to do something to prevent more gun violence two years ago I started working with a buddy of mine named Chad we want to make Hardware smarter so that it can prevent injuries and fatalities in a previous video we showed you our solution to prevent kickback of a circular saw if we detect kickback using machine learning then we stop the blade I hope to update you on that project in the near future by the way but today I'm going to tell you about something we were working on in parallel to that project about a year ago we decided to try to create a technology a box really that could be plugged in to existing security camera systems in order to detect guns where they shouldn't be we do not think we have the solution for school shootings we think we have a component that could be rolled in to a much larger strategy so basically it's a gun detector right how many months you can put video into this box and the box tells you if there's not only a gun in the frame but we're in the frame it is and draw a little box around where the gun is I'll also associate a confidence level too right right and so it's a probability of 0 to 1 mm-hmm yeah and so if it's a 1 then it knows without a doubt it's a gun we have the unit sitting here we've got the video that's coming from this webcam right here we've got your phone set up right there which is about to die by the way you need to charge your phone so check this out we've got some guns right here okay these are empty and just let me show you that because it's important so that's empty this one's also empty so we're going to do a demonstration so what I'm going to do is I'm going to show those firearms to the webcam now things to keep in mind here is it matters so when I'm going to pull this gun into frame of the webcam it's out of frame right now and I pulled in the frame the system should see it right right so I'm bringing in the frame it's detected the gun it texts you the picture of the gun you can pick up the other gun that's it that's a difference it's a discolor is it tracking it they got it they got it and so it just has to track it in what does it do this I'm a confidence level right like what happens if you have two guns right it should detect them both what I was trying to do is only detect guns that somebody's holding okay because if it's in somebody it's if it's in a holster officers here yards holster he's not carrying it there's no chance of it going off so the interaction of the gun to the hand is important yeah I trained it to look for cues that were unique to guns like the hammer like the slide and the barrel and the trigger and the trigger guard as well as your fingers near those items right man this is fascinating real quick I want to say something about what he just said Chad said he was training the system to only detect guns that have hands on them and are out and ready that's an active engineering decision and this is why that's important when we told a company that we were working on a gun detection system and we'd like to see if they were interested they asked us if we could detect guns through clothes we said no we had decided not to because we felt like that was an invasion of privacy they quit talking to us at that point because they were working on millimeter wave radar technology and microwave technology to see through your clothes now there may be a opportunity to use this type of technology in private situations in a closed environment where certain people are being protected or something along those lines however in public do we really want do we really want to live in a world where we're constantly being hit with x-ray things that look through our clothes I just want to go on record saying that's not a good idea I mean the Fourth Amendment is about search and seizure it is good for everybody no matter where you're at on the gun debate the Fourth Amendment is a good thing and we should preserve it so I just want to have this conversation now before we normalize the idea of technology that can see through your clothes being open on the street there I've been on record thanks I wanted to know what would the computer need to see in this frame in order to call us again what are the specific features on the weapon that are different from everything else in the background and so then you just trained it by playing video right yeah i pulling frames out of videos and and marking within those frames that this is a gun that this is what's important of the image of the gun not just a big box around the hole you know if somebody's holding a gun I didn't draw a big box around their whole arm I focused in on things that are unique to guns to bring the false alarm right down interesting because one thing that like the fingers on the front like if you're holding the gun those fingers are something that you're gonna see those are those are kind of unique you hold a gun in a certain way right everybody knows how to make it I think your gun yeah the other thing is you hold your cell phone like that too and other objects like that as well so I had to do not only a whole bunch of images of guns but I had to do a lot of hard- mining and finding objects that it was thinking more guns initially and then teach it that those are not guns so I did a lot of it I found a lot of YouTube videos of people with cell phones youtubes are great for teaching your computer what to do so I'm like I'm literally picking up a gun and it's detecting the gun and I'm picking up a cell phone and it's like oh it's so fun right this is like real dude so how many thousands of images do you take you put in I think there are about 30,000 images in the library right now and you went through and ID'd him yeah that's what took so long right cuz I remember like Chad would call me and his brain would be like total mush what's not dude oh not nothing much just sitting here looking at all these gun images lighting kind images oh man so seriously how many hours do you think you have highlight and gun images I would say like eighty straight hours really yeah man yeah but the work I wouldn't say that work is done but the foundation is done right like this is a jumping-off point right right I think a big key of it was that there are other gun databases out there from the movies and other you know things that people put together but I had to go to a gun store owner and have him give me permission and the people in the store permission to take pictures of them holding guns I wanted lots of different hands and you know everybody's hands are a little bit different lots of different guns different types of guns and a gun store was a really good way to do that so you're telling me that you're basically making a device to protect people against guns and you got gun stores to participate in that data collection process did you tell them what you were doing yeah and so they're all on board yeah but this is also a function of the resolution the web camera right yeah obviously it's a lot easier to detect a gun that's really close to the webcam and the confidence goes straight up through the roof when you do that like it's saying 100% it is a gun but if you backup and this is a function of computing power I would assume no it's it's a function of training how will you train the system we're on a plane by the way he's not muzzle sweep than me so it's a function of training this method is a thing it's a thing they yeah so that's interesting so the goal ultimately would be able to get a processor that's like fast enough and more powerful so that you can you could be all the way across the room here and you would still still detect see there's no way it's gonna get it there because it's against this background right well it's more there's not enough pixels on the target from that far away so it's a spatial density right this number of pixels on the target yeah super interesting but it has to be able to get it in all different aspect angles exactly right the ability to track through multiple vantage points if you will I wonder if he'll do that it won't get that now it doesn't see that they think that the things that's a phone doesn't it yeah and that was the big challenge to try to discriminate actual guns from other things you might be holding so once you've detected the gun what do you do with that I was in class today and the school that I go to has emergency plans hanging in every room so I went and got one today and I opened it up to the active shooter situation to see what it is you're supposed to do and the answer is you barricade yourself in a safe place and then you contact the authorities you call 911 and try to get first responders there and you give all this detailed information to the first responders think about an active shooter from the perspective of the 911 operator call center one thing that happens in emergency situations like this is people get confused because there's too much information some of its old some of it's not true some of the speculation that's difficult for the first responders however if you are able to real-time and see every room where there's a gun brandished whether it's a good guy or a bad guy the people on the ground would have real-time information that's actionable they could do something because they have the gods eye view of the situation and there's not a bunch of conflicting reports pouring in confusing people so where are we at engineers have created gun detectors not just us there a lot of them on the market what happens with this technology the purpose of this video was to start that discussion this technology exists it's here today but as we move forward into a world where this kind of stuff works what do we do with it we need to think about these issues so I've got a contact form linked down in the video description reach out to us if if you have a countermeasure that you're developing for active shooter if if you have a need for library information if you're trying to implement this into your security camera system just reach out also in an upcoming video here on smarter every day I plan on discussing these issues not just the detection component but the whole strategy with some people that are working on technologies to implement the detection of guns in a room - how does that information get to a first responder I learned things and I want to get that information out there because I want people to know what's possible for your school your church your business in your area all this being said the the whole goal here is to contribute to this overall discussion of how can we help save lives through uses of technology that's it that's what I want to do so I hope you found this valuable and helpful please keep your comments and discussion intelligent and respectful I would greatly appreciate that and with that I want to say thanks to the sponsor this episode of smarter every day is sponsored by hellofresh which is a milk it delivery system and what are you making tonight couscous er is it cost close I think so how does this work they send me this recipe card where it can see step by step instructions on how to make it awesome meal hello fresh is the simplest way to cook food at home this any pre-measured ingredients that no matter how comfortable you are in the kitchen you can make delicious awesome meals have you ever cooked pork chops are you intimidated by it there's something for everyone there's family recipes calories smart options or even vegetarian meals hellofresh is a really cool way to get your family around the table and eat a good meal it's flexible - like if you're going out of town you can totally pause it if you want to change the day of the week that it's delivered they will work around your schedule hello fresh is now from 566 per serving and you can get started with eight free meals that's 80 bucks off your first month Avella fresh by going to hello fresh calm and entering smarter eating you totally know what you're doing I know you use this go for it that that goes in there I think well just read the instructions don't let me derail you here super big thanks to El oppression for supporting this video to to clear all did you take mommy's roll why did you do that okay cool thanks for putting the flowers on the table those are nice
A2 初級 銃器 - スマートな毎日225 (The Gun Detector - Smarter Every Day 225) 4 1 林宜悉 に公開 2021 年 01 月 14 日 シェア シェア 保存 報告 動画の中の単語