Many sites already have cameras covering the spaces where a decision is made: parking, loading bays, entrances or work areas. Before buying new hardware or software, it is worth testing whether that footage can answer the question. These four steps run that test with clear privacy rules and a measurable result.

Step 1 — Define the question

Goal: Be precise about what the cameras should tell you.

  1. Write the question as a decision: for example, which spaces are free, or when a bay is occupied.
  2. Name who will use the answer and how often.
  3. Decide what accuracy the decision actually needs.
  4. Agree privacy rules: no identification of people, retention limits and who can see footage.

You finish with: A one-page brief with the question, users, accuracy bar and privacy rules.

Step 2 — Audit the views

Goal: Check whether the footage shows what matters.

  1. List the cameras covering the area and their angles.
  2. Pull sample footage across times of day, weather and busy periods.
  3. Mark views where the target is blocked, too small or poorly lit.
  4. Note where processing would happen, on site or in the cloud.

You finish with: A camera list with usable and unusable views.

Step 3 — Test on real footage

Goal: Measure results before committing.

  1. Label a sample of frames by hand as the reference answer.
  2. Run a pre-trained or lightly adapted model on the same frames.
  3. Compare results with the reference and record where it fails.
  4. Review failures: camera position, lighting or model limits.

You finish with: A measured accuracy result with known failure cases.

Step 4 — Decide what to change

Goal: Choose the smallest change that meets the accuracy bar.

  1. Keep the cameras that already meet the bar.
  2. Reposition or relight before replacing hardware.
  3. Add new cameras only where the test shows a clear gap.
  4. Plan how results reach the people or systems that use them.

You finish with: A recommendation with the minimum hardware change and next step.

Common pitfalls

  • Buying cameras before testing the ones you have.
  • Testing only on daytime footage.
  • No privacy rules agreed before collecting footage.
  • Chasing accuracy the decision does not need.

See it as a workflow

Facilities & operations

Parking and space occupancy from cameras

Use existing cameras to count which spaces are free or occupied and feed the count to signage, an app or a dashboard, with people checking anything unusual.

See the workflow

More guides like this

Related answers

Read as Markdown Build the missing piece