Computer Vision · Movement analysis
Movement Recognition with AI: Use in Rehab, Sport and Workplace Safety
Whether a rehab exercise is done correctly, whether the knees are aligned in a turn, or whether a lifting movement strains the back: an expert sees it when standing next to you. AI movement analysis makes that observation possible when nobody is watching.
Julia Rose
Key takeaways
- AI detects joints such as shoulders, knees and hips in camera images, builds a skeleton model and measures angles and movement sequences.
- The movement is compared with criteria set by experts, and the system gives feedback immediately.
- In a research partnership, a 3D coach checks rehab exercises at home in real time, without a physiotherapist in the room.
- Processing on the device helps with privacy: skeleton points are processed further instead of video images.
How AI movement analysis works
- Capture: one or more cameras record the movement, in 2D or in 3D with depth cameras.
- Detect the pose: a model finds the joint points in every frame, such as shoulders, elbows, hips, knees and ankles, and connects them into a skeleton model.
- Measure: the points yield joint angles, posture, range of motion and timing.
- Assess: the system recognises which exercise or movement phase is running and compares it with criteria defined by experts.
- Give feedback: deviations are shown immediately, such as a knee collapsing inward or a back that is too rounded.
3D capture from several cameras provides the spatial accuracy needed for joint angles and range of motion in rehabilitation. For many sport applications, 2D video is enough.
Applications
| Area | What AI takes over |
|---|---|
| Rehabilitation | recognising exercises at home, assessing them against the therapists’ criteria and giving corrections |
| Sport and training | analysing technique, for example knee angles in skiing, the shooting motion in basketball or the swing in baseball |
| Fitness | counting repetitions and checking posture during the exercise |
| Workplace safety | spotting straining postures and risky movements, such as when lifting, and improving ergonomics |
| Equipment | tracking equipment as well as the body, such as ball flight or ski position |
More applications for sport and fitness companies are on our AI for sport and fitness page.
In practice: rehab at home
In practice: 3D movement analysis
After surgery and in older age, patients often have to do their exercises alone at home, and done wrong they can do more harm than good. In a research partnership with hospitals, universities and technology providers, we developed a 3D coach: several Azure Kinect cameras capture the whole body, pose estimation recognises the exercise and checks it against the physiotherapists’ criteria, and corrections come in real time. VR game elements help patients keep going.
What organisations gain
- Consistent feedback: every repetition is assessed against the same criteria.
- Support without presence: exercises and training can be checked when no expert is on site.
- Immediate correction: mistakes are caught before they become habits.
- Measurable progress: range of motion and technique can be compared over weeks.
The assessment does not replace the experts. Therapists, coaches or safety officers set the criteria and decide on consequences.
Building privacy in technically
Camera images of people are sensitive, especially in rehabilitation and at work. A lot can be done technically:
- Processing on the device: the analysis runs locally, and video images do not leave the room.
- Skeleton points instead of images: only joint positions and measurements are stored and analysed.
- Measure only what is needed: the system measures only what the task needs, such as the angles of an exercise.
- Align early: data protection officers and, in companies, employee representatives are involved from the start.
How to get started
- Define movement and criteria: which exercise or task should be assessed, and how do experts recognise correct execution?
- Clarify the recording setup: number and position of cameras, 2D or 3D, lighting and environment.
- Prototype with real recordings: compare the results with expert assessments.
- Integrate: show feedback where it is needed, in the app, on a screen or in VR.
How AI video analysis is used beyond this is covered in our article Video analytics with AI.
Assessing movement automatically?
We work out with you which movements can be captured and how, which cameras are needed and what a first prototype looks like.
Request a process analysisFrequently asked questions
A model detects joint points such as shoulders, knees and hips in camera images, connects them into a skeleton model and measures angles, posture and movement sequences. These are compared with defined criteria.
For example in rehabilitation for exercises at home, in sport and fitness for technique and posture, and in workplace safety to spot straining postures and risky movements.
Standard cameras are enough for many applications. Where spatial accuracy matters, such as in rehab, depth cameras are used, often several for 3D capture.
Not necessarily. The analysis can run directly on the device, and only skeleton points and measurements are processed further.
Julia has been part of theBlue.ai since 2019 and has accompanied the development of AI applications in the enterprise environment since the company’s early days. In her role as Marketing Lead, she works closely with the engineering and consulting teams and makes complex technical topics understandable and accessible for decision-makers.
In her articles, she writes about practical experience from enterprise AI projects, as well as the challenges and opportunities of using AI in companies.