How to film a workstation for an ergonomic assessment

A video-based ergonomic assessment is only as reliable as the video behind it. The risk methods (OCRA, NIOSH, REBA) do not check what they are fed: a clip that shows the wrong operator, half a cycle or only the heaviest variant of a job produces a score that looks exactly like a legitimate one.

This guide collects the field rules that decide whether a recording can be trusted. They come from the assessments Postura runs every week on production lines, and from the questions customers ask most often when they start filming.

It is written for the people who organise and record the videos: HSE managers, ergonomists, line leaders and the colleague holding the phone. You do not need to know the risk methods to follow it.

What the guide covers

• Before you film: the documents to prepare and how to explain the recording to operators.
• Who and what to film: the operator, the cycle and the conditions the video has to represent.
• How long to record: the recommended length for each kind of task.
• Framing and camera handling: what the analysis needs to see and what breaks it.
• Checks before you upload: the mistakes that are invisible in the result and how to catch them.

The one rule behind all the others

The assessment describes the workstation as it normally runs. Every choice in this guide, from the operator you film to the minute you stop recording, follows from that.

Settle the paperwork and the communication before the first recording, not after: plan for one to two weeks of lead time. Filming people at work involves data protection rules and, in many plants, the worker representatives.

The documents to prepare

• Data protection impact assessment (DPIA): under Article 35 of the GDPR, an impact assessment is required before processing that is likely to result in a high risk to people; supervisory authorities, Italy’s Garante included, list systematic monitoring of employees among such cases. Share the vendor’s documentation with your Data Protection Officer at the start, so they can adapt it to your context.
• Notice to each worker filmed: every operator who is recorded receives a notice explaining how their image and the biomechanical parameters extracted from it are processed.
• Agreement with worker representatives: where the law or the plant’s industrial relations require it, agree the recording campaign with the unions before it starts. In Italy, for example, Article 4 of the Workers’ Statute (Law 300/1970) allows audiovisual equipment from which remote monitoring of workers may derive only for organisational and production needs, workplace safety or the protection of company assets, and after a collective agreement with the union representatives or, failing that, an authorisation from the Labour Inspectorate.

Requirements differ by country and by company. Treat this list as the starting point for your DPO and HR, not as legal advice.

The message for operators

The key sentence is simple: we are analysing the workstation, not the worker. The movements an operator makes depend on how the work is designed, not on personal choices. Saying it clearly, before the camera comes out, removes most of the nervousness.

Two more points help:

• The analysis protects the operator as well as the company: an up-to-date risk assessment is the basis for fixing the stations that hurt.
• The videos can be used for training, to share the way of working that loads the body least.

Privacy by design

In Postura, faces are obscured irreversibly during processing: people in the frame are recognised and their heads covered, and no unobscured original is kept. The obscuring is calibrated to cover too much rather than too little; if you spot a frame where a face is visible, report it to support. Videos are kept for ten years on a dedicated server in Italy. Say so when you present the campaign: it answers the first question most operators ask.

Film the typical trained operator doing the standard cycle during normal production. A video that departs from any of the three gives a score for a job that does not really exist, too high or too low.

Who to film

Operator	Film?	Why
Trained operator at normal pace	Yes	Knows the standard cycle and runs it with realistic efficiency.
Trainee	No	Uncertain movements, slow pace and errors are not the station’s real load.
Veteran with decades on the same job	No	Personal shortcuts lower their effort but cannot be standardised.

The same logic applies to handedness. If the station is normally worked by right-handed operators and the person available on the day is left-handed, the recorded cycle is not the one the station usually runs. Reschedule the recording rather than noting the mismatch afterwards: it is invisible in the result. Individual cases belong to health surveillance and the occupational physician.

What to film

• The standard cycle, as written in the procedure. Ask the operator not to speed up for the camera.
• No out-of-standard gestures. If the operator plays with the screwdriver between pieces or adds steps that are not in the procedure, that is not the cycle. Flag it to the supervisor as a training point instead of scoring it.
• Normal production, not end of shift, product changeover or extraordinary maintenance.

Record at least two or three complete cycles for a task with a takt time, one to four minutes for repetitive work without one, and every phase for a multitask workstation. Videos that are too short or too long both degrade the result.

Type of task	Recommended length
Cycles with a defined takt time	At least 2–3 complete consecutive cycles
Repetitive work without a takt time	1–4 minutes of typical actions
Multitask workstation (several phases)	Long enough to cover every relevant phase

Why length matters

• Too short: processing can fail or give an imprecise analysis. Short videos often need a manual quality review before the result is released.
• Too long: processing takes longer and the video fills with noise, such as actions that are not representative, meaningless pauses and setup.
• Two or three cycles let you check the variability between cycles and confirm that the pattern of actions is stable.
• Without a takt time, the video length is the observation window. One to four minutes usually captures the pattern; use judgement for more complex tasks.

Keep the machine waiting time

On tasks with a machine cycle, such as a press, do not cut the time the operator waits for the machine to finish. Keep it in the video and ask the operator to stay still rather than gesture. The system then recognises it as time without technical actions: it lowers the action frequency and, when it lasts at least 10 consecutive seconds in every cycle, counts as recovery inside the cycle. Cutting it makes the work look denser than it is.

Keep lengths consistent

When you record several videos of the same station, keep their lengths similar. Very uneven lengths in the same batch cause comparison and processing errors. The simplest fix is an internal standard, written down and shared with everyone who films.

Keep the operator fully in frame for the whole video, from the first second. That is the one hard requirement: the camera can move, the background can be busy, but the operator must stay visible.

How the AI finds and follows the operator

• At the start of the video, the tracking locks on to the largest person in the frame. Start recording with the operator you want to assess close to the camera and clearly visible.
• Once locked, the tracking stays on that person even when colleagues walk through the scene.
• It is lost only if the operator leaves the frame completely.

If the operator is not in the frame when the video starts, the analysis cannot pick the right subject. Start the recording on the operator, not on the workstation.

Moving the phone is fine

Moving the camera to follow the operator, even quickly, does not penalise the analysis. The system is designed for dynamic recording, which is how real workstations are filmed. Walk with the operator if the job moves along the line.

What matters is continuous visibility. For OCRA, Postura asks for:

• Two videos per task, one from the right side and one from the left side; a video of the hands is optional.
• Hips to head in frame at all times, with the head and pelvis visible.
• 720p to Full HD at 30 fps, not 4K.
• The same length for the right and left videos: very different durations block the upload.

Practical tips

• Prepare the shot before you press record. If unlocking the phone and placing the operator take up a large share of the clip, the video is mostly setup. A dedicated person filming, or a gimbal with tracking, solves it.
• Film in normal light, without strong backlight from windows behind the operator.
• One task per video. Record the right and left side as two separate videos rather than one that switches between them.

Before uploading, check that the video contains exactly one complete cycle, or the number of cycles you declare, and that the declared cycle time matches it. A wrong cycle time silently changes everything computed from it.

The recorded segment is not the cycle

This mistake is invisible in the result, and it comes in two opposite forms.

• The clip contains more than one cycle. On a station that handles pieces that are not identical, a clip covering two of them and declared as one cycle doubles the declared cycle time. Action frequency and the index follow it.
• The clip contains less than one cycle. If the clip stops before the cycle closes and leaves out refills, packaging disposal or any step that happens once every few pieces, the analysis scores the dense part of the work as if it were the whole. The station comes out red for a reason that is in the recording, not in the job.

The check is the same in both cases: find the first and last frame of one complete repetition and confirm that the declared cycle time is the duration of that repetition, not of the clip. Where pieces differ, agree with the people who run the station which one the cycle represents.

A practical corollary: when someone disputes a score, check the cycle time first. They are often disputing a cycle time without knowing it.

Pick one workstation, apply the checklist in chapter 5 and film it this week. One good video teaches more than a campaign planned on paper: it shows you how long your cycles really are, which conditions the station runs and what your operators need to hear before the camera comes out.

With Postura

Upload the videos from your phone and you get both readings of the same recordings: the ergonomic risk score (OCRA from the video, REBA where enabled on your account, with NIOSH and push-pull from the measurements you enter) and the time and methods analysis of the cycle. The first videos you upload are reviewed by our team, with feedback on the recording itself.

• See it on your own footage: book a demo.
• Choosing the method: read OCRA vs NIOSH vs REBA.
• Reading the OCRA score: read OCRA High-Precision Checklist.