Skip to main content

Record video for an AI time study

Camera placement, operator identification and other tips for accurate AI time studies.

Written by Angad Singh

Record video that gives accurate AI time studies. Good footage is the biggest factor in how accurate your study is.

Camera placement

  • Overhead or high-angle views work best, but keep the camera close enough that operators stay large in the frame. If people look small, the AI struggles to track them.

  • Record in landscape.

  • Keep the whole work area in view and avoid blind spots, such as operators working behind walls or machines.

  • Use good, even lighting and avoid strong backlight, such as bright windows behind the operator.

  • Use one camera per workstation or section.

  • If you use several cameras, start them at about the same time.

Important: Avoid overlapping camera views of the same work, because the same activity can get counted twice.

Identify operators

  • Give each operator a large, high-contrast number on the front and back, or a different colored vest.

  • Make sure numbers aren't hidden by jackets, aprons or welding gear.

  • Keep other foot traffic through the area to a minimum.

  • Use just the digit, large and bold (for example "7", not "Number 7"), so it can be read from above or at an angle.

  • Optional: at the start, have operators face the camera for a few seconds.

During recording

  • Record the process as normal. Operators shouldn't talk to or pose for the camera, because the AI may count it as waste.

  • Turn on the camera's date and time overlay if it has one.

  • Action cameras can overheat on long recordings, so keep them ventilated.

  • Check the memory card has enough free space for the whole recording before you start.

Video length

  • Each study can be up to about 4 hours of video. Longer recordings go in as separate studies that you compare.

Tip: For a brand-new process, a shorter clip of about 10 to 15 minutes is a good first study to review and certify.

File size and resolution

  • Record at 1080p. 4K files are much larger and slower to upload, and don't give better results.

  • On a slow or unstable connection, split long recordings into parts of about an hour and upload each part as its own study. Use Compare to view the parts together.

  • Very large files, such as tens of gigabytes, take a long time to upload and are more likely to fail on a slow or unstable connection.

  • LeanSuite compresses the video after upload and keeps a copy, so you can delete the file from your computer once the upload has finished.

Note: You can't import a video directly from a camera's cloud storage. Download the file to your computer first, then upload it.


Related articles

Did this answer your question?