All projects

Build sheet · 2024

Autonomous Surveillance Robot

I designed and built this robot from scratch. It patrols on its own, spots people in live video and sends an alert, with everything running on a Raspberry Pi on the robot itself.

  • Python
  • Raspberry Pi
  • OpenCV
Render of a small black tracked robot with a front camera and two antennas
Fig. 1 · a render of the robot. The real one is below.

Fig. 2 · the real robot

The robot

Photograph of the finished robot on a tiled floor: a black box body between two tank tracks, a webcam in the front panel and two antennas at the back

  1. Camera

    A webcam at the front is the robot's only eye. Every decision starts from its video.

  2. Tracks

    Two motor-driven tank tracks, one each side, steer the robot by running at different speeds.

  3. Body

    Houses the Raspberry Pi that runs the detection, and the electronics that drive the motors.

  4. Antennas

    The wireless link the robot uses to send an alert when it sees someone.

Fig. 3 · control loop

How it works

The robot runs the same loop for as long as it is switched on. Everything happens on the robot itself, with no cloud and no graphics card, just a Raspberry Pi the size of a credit card.

Raspberry Pi, on board

  1. See

    The camera streams live video to the Raspberry Pi.

  2. Detect

    OpenCV looks for a person in every frame.

  3. Decide

    Alert logic decides whether what it saw is worth raising.

  4. Act

    The motors move the robot, and an alert goes out.

Notes

Challenges

A Raspberry Pi has only a small fraction of a laptop's computing power, and the video never stops coming in. The detection had to be light enough to keep up, and the alert logic had to be tuned carefully so it didn't send an alert for every shadow.

A tracked robotic system integrating perception, decision-making and actuation.

  • Real-time human detection and event-based alert logic running on a resource-constrained Raspberry Pi: edge deployment of a live video inference pipeline
  • Sensor, motor control and communication integration, tuned to reduce false triggers