Backpack setup

🚨 Cyclops: AI-Powered Backpack Sentry

Inspiration

With the rapid advancement of computer vision and hardware capabilities, we wanted to tackle a problem that directly impacts personal safety. We noticed a common issue among travelers and everyday pedestrians who carry backpacks — lack of awareness of what’s happening behind them.
There are numerous situations this system could help prevent, such as stalking, pickpocketing, or unwanted proximity. Our goal was to build an awareness system that leverages AI to enhance safety and peace of mind in everyday movement.

What It Does

Cyclops is an AI-powered awareness system that mounts a rear-facing camera onto a backpack to monitor the user’s blind spot.
We developed algorithms and criteria to identify “suspicious” behavior, such as:

A person remaining in frame for an extended period
Someone moving closer over time
Repeated reappearances of the same individual

If the system deems a person suspicious, it alerts the user’s phone via push notification, providing real-time situational awareness.

How We Built It

We used computer vision models like YOLOv11 and DeepSORT for image and video stream processing.
Alongside these models, Python scripts were used to extract data and generate statistics, which were then fed into GenAI models and our own algorithms to determine if a person was suspicious.
Finally, we integrated Firebase and Android Studio to create a real-time mobile notification system that connects the CV pipeline to the user’s device.

Challenges We Ran Into

Computer vision is a notoriously complex field. Our biggest challenge was the re-identification (ReID) problem — determining whether a person leaving and re-entering the frame is the same individual or a new one.
This problem is complicated by factors such as:

Image resolution affecting embedding quality
Varying lighting conditions
Changes in pose or orientation
Distance and occlusion

Although our model isn’t perfect, it serves as a proof of concept that demonstrates the potential of our full system.

Accomplishments We’re Proud Of

We’re proud of building something unique and interdisciplinary. Successfully merging multiple technologies was a huge accomplishment for our team.
Our tech stack includes:

Computer Vision (YOLOv11, DeepSORT)
Arduino hardware, ultrasonic sensor, webcam
Python scripting and GenAI integration
Firebase database
Android mobile app development

Bringing all these together into a cohesive prototype was no small feat.

What We Learned

We learned a lot — both from our successes and our limitations.
While we mastered the tools in our tech stack, we also learned about project scope, integration challenges, and adaptability.
Originally, we attempted to create a fully autonomous agentic system using Google’s ADK. Although we pivoted away from that approach due to hardware limitations, we gained valuable insights from the process.

What’s Next for Cyclops

Looking ahead, we plan to:

Improve computer vision performance using newer models and optimized quantization techniques
Enhance the mobile app with more safety features and user customization
Streamline the workflow for better efficiency and maintainability
Explore on-device inference for low-latency and offline operation