Skip to content
Charles Walsworth
All experiments
Experiment

Testing Object Detection on a Raspberry Pi (Sample)

Experimenting with real-time object detection on inexpensive hardware to see where the frame rate actually falls over.

  • Python
  • OpenCV
  • Raspberry Pi

Sample content. This entry is placeholder material included to demonstrate the layout, not a record of real work.

The question

How much object detection can a Raspberry Pi actually do in real time before the frame rate becomes useless?

Setup

  • A single USB camera at 640×480.
  • A small pre-trained detection model, run per frame with no tracking.
  • Frame times measured over a fixed two-minute window rather than eyeballed.
python
import time
 
frame_times: list[float] = []
 
while capturing:
    start = time.perf_counter()
    frame = camera.read()
    detections = model(frame)
    frame_times.append(time.perf_counter() - start)

What I expect to find

That preprocessing, not inference, is where a surprising share of the time goes — and that resizing on the camera rather than in Python moves the number more than swapping models does.

Where this could go

If the frame rate holds up, this becomes the perception half of a small robot. If it does not, the interesting question becomes what has to be given up first: resolution, frame rate, or model size.