Cameras

A camera turns your robot from a blind machine into one that can see the world. That single upgrade opens up a huge range of possibilities — from following a coloured ball to recognising faces to navigating a room without bumping into the furniture.

What is a Camera (in robotics)?

In robotics, “camera” covers everything from a basic USB webcam to a dedicated board camera like the Raspberry Pi Camera Module 3. The Pi Camera connects directly to the CSI port on a Raspberry Pi, giving you full resolution control and access to features like autofocus and HDR. The newer Raspberry Pi AI Camera adds a built-in neural network accelerator (Sony IMX500 chip), letting you run object detection models right on the sensor — no extra processing needed from the Pi itself.

Resolution matters, but frame rate matters just as much for a moving robot. For most beginner projects, 1080p at 30 fps is plenty — and even 640×480 is enough for real-time colour tracking.

Libraries like OpenCV give you the tools to process what the camera captures: detect edges, find contours, track objects by colour, or feed frames into a machine-learning model. CVZone builds on top of OpenCV and makes common computer-vision tasks (hand tracking, pose estimation, face detection) accessible with just a few lines of Python.

The vision pipeline

Raw pixels on their own don’t steer a robot — they have to be turned into a decision. Almost every camera robot follows the same chain: capture a frame, process it to find something meaningful, then act on what was found:

flowchart LR
    CAM["Camera<br/>(captures frames)"] --> FRAME["Frame<br/>(image array)"]
    FRAME --> PROC["OpenCV / model<br/>(find object,<br/>face, colour…)"]
    PROC --> DECISION["Decision<br/>(where is it?)"]
    DECISION --> ACTION["Robot action<br/>(turn / follow)"]
    ACTION --> CAM

That final arrow back to the camera is the important one — it’s a loop. The robot acts, the scene changes, the next frame comes in, and round it goes many times a second. That’s what lets a robot smoothly follow a face or a coloured ball rather than just spot it once.

Why robot builders care

Vision is the richest sensor a robot can have. A single camera gives you colour, distance cues, shape recognition, and motion detection — all at once. Compare that to an ultrasonic sensor, which only tells you there’s something 30 cm ahead, or an infrared sensor, which just sees a line on the floor. A camera tells you what is ahead, not just that something is there.

That richness does come at a cost: processing video in real time takes real compute, which is why the Raspberry Pi 4 or 5 is a natural pairing. For lighter tasks, an ESP32-CAM module adds basic streaming video to a tiny board for just a few pounds.

Get started

The best way in is to build something with immediate visual feedback. Kevin’s Computer Vision on Raspberry Pi with CVZone course starts from a basic camera feed and works up to posture detection and hand-tracking — no computer-vision background needed. If you want to go further and train your own models, Building Object Detection Models with Raspberry Pi AI Camera covers PyTorch training and edge deployment on the AI Camera.

For inspiration on what a camera-equipped robot actually looks like as a finished project, the PIKON Camera build shows a high-quality Raspberry Pi camera enclosure built from scratch — and the Face Detection with Trilobot post puts it all together with a robot that follows a detected face.

Start with a still image, then a live feed, then try tracking a coloured object. Each step builds confidence, and before long your robot won’t just react to the world — it’ll actually see it.