Math Bridge: Camera Sampling Before Edge AI

← Back to Edge AI & ML Applications
Math BridgeEdge & FogStruggle-friendly runway

Can a camera preserve motion before the model sees it?

Follow one packing-line vibration through frames, aliases, pixel codes, and the later model-compression step.

Edge Eddie, the guideEdge Eddie guides
The one targetKeep sensor rounding separate from model rounding.
The chapter case30 fps, 22 Hz motion, illustrative 10-bit ADC.
What it buys youAn input record that exposes lost motion before inference.

A field team faces an unresolved physical question: Can a camera preserve motion before the model sees it? They must answer it before changing camera frame rate in frames per second on the real device. Predict the direction first.

See the relationship before changing it

The figure reads from left to right. The blue card is camera frame rate in frames per second. The middle card applies this page's relationship. The green card is real motion. Walk the arrows once: set the input, apply the rule, then read the result with its unit.

The retained audit below checks several chapter fixtures. This added model holds every other chapter fixture fixed, so the numeric fixture does not switch without explanation.

Camera frame rate in frames per second changes real motion An input card leads through the page relationship to the real motion result. SET INPUT ONE CONTROL APPLY RULE predict calculate check units READ RESULT
Walk the arrows. The slider changes when frames are taken. It does not change how brightness is rounded or how model weights are stored.

Derive the baseline in four named moves

  1. 1

    Name the input. The chapter baseline for camera frame rate in frames per second is 30.

  2. 2

    Name the relationship. fNyquist=fs/2; falias=|nfs-f|; q=100/2^N; eRMS=q/√12

  3. 3

    Substitute the chapter fixture. Set camera frame rate in frames per second to 30. The page ledger gives real motion as 22.00 Hz.

  4. 4

    Read the result. Keep Hz beside the value. Use it only inside the technical boundary on this page.

Predict, then change camera frame rate in frames per second

Try Predict the direction of real motion. Move one control, calculate, then check your prediction.

30
Chapter baseline
Real motion

Observe The slider changes when frames are taken. It does not change how brightness is rounded or how model weights are stored. Reset the control to 30 and compare real motion.

Explain Only camera frame rate in frames per second moves here. The other chapter fixtures remain fixed.

Check yourself

What should you do before trusting a moved-control result?
Answer: Predict its direction, apply the shown relationship, keep the units, and reset to the worked baseline.
What does this small model leave out?
Answer: Only camera frame rate in frames per second moves. Field effects named in the page's technical boundary stay fixed.

Honesty note. This teaching model changes only camera frame rate in frames per second. It replays the chapter fixture; it is not a field prediction.

1. A frame is one look in time

A camera does not watch continuously. At 30 frames per second, it takes 30 separate looks each second. A repeating motion can move between those looks and seem slower than it is. The pixel ADC then rounds brightness into codes. Model compression happens later and changes stored weights, not the captured scene.

Edge Eddie: Ask what the camera preserved before asking what the model learned.

2. Name every algebra move

1

Halve the frame rateThe highest unambiguous motion rate is fs/2.

2

Find the nearest sampled copyRound f/fs to n, then take |nfs − f|.

3

Count pixel codesN bits provide 2^N levels.

4

Divide full scaleq = 100% / 2^N.

5

Convert a uniform step to RMSeRMS = q/√12.

3. Work the chapter numbers

30/2 = 15.0 Hz; |30 − 22| = 8.00 Hz; 100/1024 = 0.0977%

The chapter's camera cannot identify 22 Hz motion correctly at 30 fps. With an illustrative 10-bit pixel ADC, the ideal RMS rounding floor is 0.0282% of full scale and the ideal quantisation ceiling is 62.0 dB.

4. Try one controlled change

fNyquist=fs/2; falias=|nfs−f|; q=100/2^N; eRMS=q/√12

TryMove only the camera frame rate. The 22 Hz motion and illustrative 10-bit pixel ADC stay fixed.

Nyquist limit
Real motion
Motion appears at
Pixel levels
ADC step
RMS floor
Ideal ceiling

ObserveAt 30 fps, the Nyquist limit is 15.0 Hz and 22 Hz appears at 8.00 Hz. Raising the rate to 50 fps preserves 22 Hz, while the ADC readouts do not move.

ExplainThe slider changes when frames are taken. It does not change how brightness is rounded or how model weights are stored.

Technical boundaries.

This is ideal uniform sampling and quantisation.

Exposure
Motion blur and rolling shutter can damage an otherwise adequate frame rate
ADC
Sensor noise, clipping, gamma, and compression can exceed the ideal floor
Model
Passing input checks does not prove detection accuracy or safe stopping

Measure the actual camera timing, optics, pixel path, and representative line motion.

5. Keep the two rounding stages apart

The pixel ADC rounds a voltage before an image exists. INT8 model conversion rounds trained weights after training. Both use discrete codes, but they have different units, causes, and evidence.

6. Carry the evidence into deployment

Record frame timing, exposure, motion band, pixel format, ADC depth, compression, preprocessing version, model artifact, decision latency, and the safe action when input quality fails.

7. Check yourself

Why is the limit 15 Hz at 30 fps?
Answer: Nyquist halves the sample rate: 30/2 = 15.
Why can 22 Hz look like 8 Hz?
Answer: The nearest sampled copy is |30 − 22| = 8 Hz.
Can later INT8 conversion restore that 22 Hz motion?
Answer: No. It changes model weights after the camera has already sampled the scene.
Honesty boundary.

The 30 fps line belongs to the chapter; 22 Hz and 10 bits are clearly labelled illustrative values.

30 fps
Chapter camera rate
22 Hz
Illustrative machine vibration
62.0 dB
Ideal 10-bit quantisation ceiling only

Go deeper in the chapter, then test the installed camera and real fault motion.

Return to Edge AI & ML Applications with the maths unpacked →