iGotYourBack | Devpost

Inspiration

Approximately 1 in 5 women and 1 in 71 men will experience sexual assault or rape in their lifetime according to a 2010 study conducted by the Centers for Disease Control. During that same year, 65,000 individuals were reported missing and in danger indicating they were likely kidnapped. Even more disturbing, these assaults are most often committed in public spaces at night when the victim is alone and most vulnerable. Awareness of your surroundings is, of course, one of if not the most important factor in avoiding becoming a victim, but you can't always be so vigilant. This raises the question: who will watch your back when there is no one else there to?

Introducing iGotYourBack: an intelligent wearable proximity alarm

How It Works & What it does

Version 1.0 of iGotYourBack features a sonar and camera that attach to your backpack. These devices integrate with our app to first identify potential threats using a human detection OpenCV script interfacing with the camera then approximate their distance using the sonar.
When a potential threat is sensed within a specified distance, your phone and smart watch will continuosly vibrate to warn the user of this potential threat. This sets the proximity alarm to the "ready" position in which abnormal agitiation of the accelerometer in any device integrated with the app including the phone itself will trigger the alarm.
When the alarm is triggered, a noise is played through the speakers at the phone's highest audio level, GPS location is pinned and set to actively track, and the authorities are notified.
To mitigate false alarms, there is a small window when the alarm is triggered where the user can deactivate it by using both facial recognition and inputting a PIN number.

How I built it

For the prototype we used a pi to run the OpenCV application and get data from the sonar. The app mockup then shows how the when a person is detected too close to you the application sets off a panic button.

Challenges I ran into

Initially we had trouble finding a lightweight human detection and tracking algorithm that can work efficiently on the Pi， so we had to go through some testing. During the real world testing with the camera we experienced some trouble with detecting in low light situations, so we had to tune some parameters in our algorithm.

Accomplishments that I'm proud of

Proud that we were able to deploy an efficient algorithm for our detection system and produce usable results in low light situations. Successfully provided a demo of our project, along with creating an app mockup for our project.

What I learned

We learned about machine learning, including using computer vision for human detection and tracking, along with tuning parameters and writing code to achieve the best results. We also learned to develop a prototype UI for our product.