SafeWalk

Inspiration

As a group of female college students, we often have to walk back from classes or study spaces around campus late at night. These late-night walks are nerve-racking and dangerous, particularly coming from a campus in the city. We wanted a product that we knew would have our back. A product that could easily spot danger and warn us before anything terrible happens.

What it does

Our product is a backpack clip which is constantly on the lookout for potential threats behind you. It detects incoming people and objects, snatches a photo, then performs AI threat analysis. Depending on the level of threat, our project acts as an AI agent, making an informed decision on whether or not to raise any alarm, alert the user of a potential threat through audio notification, alert emergency contacts, or call emergency contacts.

How we built it

Contained inside this backpack clip is an ultrasonic sensor. When an object is detected within a certain distance (can detect up to 8 meters), this triggers the webcam to turn on and snap a photo of the surrounding area. This photo is then sent to Gemini for safety analysis. Gemini has been prompted to assess the danger level on a scale of 0-100, and to process whether the danger is coming from the left or right. The agent’s course of action depends on this danger assessment. Less than 15 and it will not do anything, between 15 and 40 it will also alert the user through earbuds what side the person is on, 40-80 and emergency contacts will also get notified, then greater than 80 and the emergency contact will be called.

Challenges we ran into

Difficulty deciding how to play audio. The first time we tried to connect to bluetooth headphones via ESP32, then discovered it was too complicated. After that we switched over to using the app to play audio.

Another challenge was learning how to integrate all the different pieces. The app, the working model, the sensor — we got them all to work. Then we had to figure out how to integrate all these pieces together.

Accomplishments that we're proud of

Any time we got a small piece of the project working was a huge success! Every time we got a valid sensor reading or got the image recognition to work, it was incredibly rewarding. On the app side, getting maps to render was especially rewarding because it was a completely new API we had never worked with.

We also feel accomplished in the fact we stuck to our original theme of safety, and built a product that could genuinely keep people safe.

What we learned

We were working with a lot of different languages - C++, python, Java, and everything had to interface together. We also learned a lot about the Google Cloud by using Google Maps API, Gemini, Firebase functions, and Firebase database.

What's next for SafeWalk

We will work on improving latency. As a safety tool efficiency is key. We also did not have access to a webcam, so in the future we would train our own model, run it on raspberry pi, and integrate it with a camera.

Furthermore, we plan on integrating haptics in the form of left and right motors to alert the user about which side the attacker is coming from.

An additional feature is saving/storing images and streaming to your emergency contacts. This will allow the app the store evidence, holding culprits accountable. It’s would also be useful to store the image and send it to the user so they can see what’s behind them.

Lastly, we would print the aesthetic case we designed to house or electronics.