Herm

Herm is an AI-powered dashcam that detects break-ins and stolen plates using ML. View live feeds, mark your car stolen, get alerts if it’s spotted, and receive warnings about nearby vehicle thefts.

Comment

Inspiration

During brainstorming for Hack Canada 2026, we started looking at problems Canadians deal with every day. One statistic stood out to us: a car is stolen in Canada roughly every 7 minutes, with many cases happening in Ontario and Quebec. Even worse, many vehicles are shipped overseas or stripped for parts before they can ever be recovered.

We realized that most current solutions — alarms, trackers, or AirTags — are reactive and easy for thieves to disable. That made us ask a simple question:

What if cars could watch out for each other?

That idea became Herm, a system that combines a smart dash device with a web platform to detect break-ins and create a crowdsourced network that can spot stolen vehicles in real time.

What it does

Herm is an AI-powered anti-car theft system built around a dash-mounted device and a web dashboard.

The device includes a camera and sensors that monitor the vehicle for suspicious activity such as forced entry or tampering. If something unusual happens, the system immediately alerts the owner.

Herm also scans nearby license plates using computer vision. If a detected plate matches a vehicle that has been reported stolen, the system flags it and reports the sighting through the platform.

Key features

  • Live dashcam feed accessible from a web dashboard
  • Break-in and suspicious activity detection
  • License plate recognition for stolen vehicle detection
  • Alerts when your car is tampered with
  • Notifications when a stolen car is spotted by another device
  • A network of devices helping track stolen vehicles

If a user's vehicle is stolen, they can mark it as stolen in the dashboard. Any Herm device that later detects that license plate will notify the owner and display the location.

How we built it

Herm combines hardware, backend infrastructure, and a web interface.

Hardware

We built a prototype device using microcontrollers such as an ESP32 or Raspberry Pi, connected to a camera module and sensors. The device captures video frames, monitors activity, and sends information to the backend.

The firmware was written primarily in C++.

Backend

The backend was built using Node.js and handles:

  • device communication
  • event processing
  • stolen vehicle reports
  • real-time notifications

We used a cloud database such as Firebase or Supabase to store vehicle data, device events, and stolen vehicle reports.

Frontend

We designed our interface in Figma and built a dashboard using Next.js.

The dashboard allows users to:

  • view a live camera feed from their vehicle
  • mark their vehicle as stolen
  • receive alerts and notifications
  • view sightings of stolen vehicles on a map

AI / Computer Vision

Herm uses computer vision techniques to detect license plates from camera frames.

Conceptually, the system processes frames like this:

detected_plates = f(camera_frames)

Each detected plate is compared against the database of reported stolen vehicles:

match = detected_plate in stolen_plate_database

If a match is found, the system sends an alert and records the location of the sighting.

Challenges we ran into

One of our biggest challenges was license plate recognition. Plates can be hard to read due to lighting conditions, motion blur, and camera angles. Getting reliable detection while keeping the system lightweight enough for a small device required experimentation.

Another challenge was connecting hardware with a real-time web system. We had to figure out how devices could send data to the backend while keeping alerts fast enough to feel instant.

We also had to work within the time constraints of a hackathon while building hardware, backend infrastructure, and a frontend dashboard at the same time.

Accomplishments that we're proud of

We’re proud that we were able to build a working prototype that connects embedded hardware with a live web platform.

Some highlights include:

  • Building a functional dash device prototype
  • Creating a live web dashboard for monitoring vehicles
  • Implementing the core concept of a crowdsourced stolen vehicle detection network
  • Demonstrating how devices can detect and report suspicious activity in real time

Most importantly, we turned an idea into a working system within a limited hackathon timeframe.

What we learned

This project taught us a lot about combining different areas of technology.

We learned how to:

  • integrate IoT hardware with modern web applications
  • work with computer vision models for real-world detection problems
  • build real-time event systems for notifications and alerts
  • collaborate quickly and divide responsibilities under hackathon pressure

We also learned that solving real-world problems often requires combining many different technologies into one cohesive system.

What's next for Herm

If we continued developing Herm, we would focus on improving both the hardware and the detection system.

Some next steps would include:

  • improving on-device AI for faster and more accurate detection
  • integrating GPS tracking for more precise vehicle locations
  • connecting to public stolen vehicle databases
  • developing a mobile app for instant alerts
  • improving privacy protections for license plate scanning

Our long-term goal is to build a community-powered anti-theft network where vehicles help detect and recover stolen cars faster.

In a large network, detection probability increases as more devices are deployed. Conceptually:

P(detection) = 1 - (1 - p)^n

Where:

  • p = probability that one device detects the stolen car
  • n = number of devices in the network

As the network grows, the chances of recovering stolen vehicles increase dramatically.

Built With

Share this project:

Updates