How AI Detects Danger (When You Can’t Call for Help)

What we're doing is we're separating in two parts mainly. Well, let's say three parts, but two that I want to talk to you about first. So, audio, we're looking, we're constantly listening when you have that when you clip that badge on and you activate it for that 10-minute walk. We're looking for any audio cues that might indicate that you're in danger. Because when we did our research oftentimes like what like a typical case scenario that happens is they jump you with a with a knife in a UK case I guess in a US case that would be a gun and they tell you hey do not move do not scream do not shout give me your things right now >> and in that situation you cannot do anything well you shouldn't do anything I don't want to say cannot but like you would put yourself

in more danger >> so you are >> exactly >> so you are in a situation where you can't even notify anyone that's happening the that specific specific audio would automatically get picked up by our batch and it would know like okay there's something wrong and that we're already in that case raising the alert in the back in the back end and based on what's happening deciding like okay are we going to play this deescalation message or are we going to raise this straight away to the police. So this is what we're doing with the audio part because audio sometimes not everything is visible on the video and that's in the audio part we're looking at factors like bunch of different factors that then come together. So from emotional analysis to sentient analysis to keyword spotting to like understanding the logic behind the sentence because there's

a difference when someone says like hey like give me your wallet and someone like hey do you mind borrowing me like you know this card and no no like it's very we have to really always like look specifically how the sentence is said the the volume the pitch um for it to like give it almost like a ranking in our back end. So that's how we how we look at the audio part. And then on the other side of um we're looking always at the video as well because we're collecting evidence your entire walk that is spotting for any sort of patterns that would be dangerous. So from object uh analysis like if you spot a gun, if you spot a knife, but also like patterns of behavior that could appear threatening like you have someone running towards you in an aggressive way. That is

something that we're looking at and that is one of piece of the puzzles that is then provided together for the AI agent to make a decision what's happening.