Threat Detection AI Security System

Inspiration

our original inspiration was the ring camera and its ability to capture simple motion detection and let the user know. We wanted to expand on this idea of motion detection and take it on another level with AI. Not only did we want it to detect motion but also specific body language that is proven to be suspicious and can potentially save a persons life.

What it does

Our security system is planned to take video snippets just like how dash cams do and compares it to a AI/ML language model which is learning and acting on hundreds of videos showing theft, breaking and entering, and even assault. If the code determines the user or their property are in danger, its sounds an alarm and contacts the proper authorities.

How we built it

We started with our Raspberry Pi which was hooked up to a usb camera, the laptop for the code, and an active buzzer. We were able to take images and videos through the usb camera which outputted to the raspberry pi. We were not able to implement our own videos but we discovered a study online centering around pretrained theft detection software and we heavily referenced the data sets and code from that. Our AI/ML consisted of OpenCV a real time computer vision library and scikit-learn which was our machine learning library. We used the scikit-learn IsolationForest library specifically due to its detailed image detection capabilities which is exactly what we needed to find small differences in human body motion to determine if they are a threat or not.

Challenges we ran into

Some big challenges we ran into early in our project was figuring out an effective way to setup the hardware like the camera and what platform we should put our code on. We initially wanted to put everything on Arduino but pivoted to Raspberry Pi due to it already having python build in which made it really easy to put our code on and see everything in real time. We also didn't have a camera module that connected to the pinouts which greatly complicated things when it came to the code since there is specific software designed for camera modules on the raspberry OS. We needed to hardcode the USB camera through the terminal which took a long time to configure but after some time we finally got our first image. Another large issue we ran into was where to start with our AI/ML. We had no clue which AI or ML to use so we had some help from ChatGPT which really pushed us in the right direction but still gave us a fair share of challenges as well. It initially gave us TensorFlow to work with but it kept coming up with Memory errors which showed that the hardware we were running on was not nearly enough. We unfortunately had to go through almost 15 iterations of that code since it always gave us different errors. OpenCV was also giving us a lot of trouble since it was having an issue downloading in our version of Python and even when we updated or removed that version, it still wasn't installing so we switched out the raspberry pi for a new one and started from scratch which solved the issue. ChatGPT recommended us use scikit-learn and with a little tinkering around, that seemed to give us something with at least an output and no major error codes which was a start.

Accomplishments that we're proud of

We managed to get most of the hardware running with the exception of the active buzzer since there was no time to solder it on to the raspberry pi pins. On top of that with a whole night of coding without an ounce of shut eye, we managed to get an output for our code without any major errors after inputting a video of a burglary. This made us super proud since there were times where we were not super hopeful with anything working out but we pushed through and kept working. We had a great balance of working on hardware and software and our efficiency won out at the end. We were also proud of meeting and connecting with so many wonderful people and experiencing everything this event had to offer.

What we learned

We learned so much about how to implement AI in projects, how to time manage and multitask between hardware and software, how to present our project in an informative way, collaborating with other hackers and discovering how useful AI can truly be.

What's next for Threat Detection AI Security System

We want to keep working on this over the summer and hopefully outsource to consumers to buy eventually. Maybe finance through kickstarter and gofundme to garner interest.