AI posture detection
How Does AI Detect Bad Posture From a Camera?
How AI posture apps detect bad posture from a phone camera: pose estimation, joint-angle measurement, and the accuracy limits you should know about.
Updated
Short answer
AI posture apps use "pose estimation" - computer-vision models trained to find key points on your body (ears, shoulders, hips, knees) in a photo or video frame - then measure the angles between those points. A common one is the craniovertebral angle, which estimates how far your head sits forward relative to your shoulders; a smaller angle generally suggests more forward head posture. It's a geometry problem applied to a photo, not X-ray vision, so accuracy depends heavily on lighting, camera angle, clothing, and how consistently you set up the shot.
The core technology: pose estimation
Pose estimation is a well-established branch of computer vision (models like OpenPose and MediaPipe popularized it). Here's roughly what happens when you take a posture-check photo:
- Landmark detection. The model scans the image and predicts pixel coordinates for body landmarks - typically the ear, shoulder, hip, and sometimes knee and ankle, on a side-view photo.
- Angle calculation. Using those coordinates, the app calculates specific angles known to correlate with posture patterns:
- Craniovertebral angle (CVA) - the angle between a horizontal line through the shoulder (C7 vertebra area) and a line to the ear tragus. Lower CVA is associated with more forward head posture.
- Shoulder alignment - comparing left/right shoulder height in a front-view photo to flag asymmetry.
- Pelvic/hip tilt estimates - used less reliably from photos alone, since this often requires a side profile with visible hip landmarks.
- Pattern classification. The app compares your measured angles against reference ranges to label a likely pattern - forward head posture, rounded shoulders, uneven shoulders, etc.
On-device vs. cloud processing
Some apps run the pose-estimation model directly on your phone (on-device inference); others send the photo to a server and process it in the cloud. On-device tends to be faster and keeps your photo local; cloud processing can use larger, more accurate models but requires uploading the image. Either way, the underlying math - landmark detection plus angle calculation - is the same. If privacy matters to you, check the app's policy on whether photos are stored or deleted after processing.
Why accuracy varies
This is the part worth being upfront about, because it's easy to overstate what a camera can do:
- Lighting and contrast. Poor lighting makes it harder for the model to find precise landmarks, especially the ear and shoulder points used in CVA.
- Clothing. Loose or baggy clothing obscures the shoulder line and can shift the measured angle.
- Camera angle and distance. A phone that isn't level, or a shot taken from a slightly off angle, changes the apparent geometry - this is why apps that guide you to a fixed distance and angle (often with an on-screen silhouette) tend to be more consistent.
- Single-plane limitation. A single side-view photo only captures posture in one plane. It can't see rotation or asymmetries that would show up from a different angle, and it captures one moment, not your posture through the day.
- No skeletal or nerve information. The model is measuring surface landmarks, not bones, discs, or nerves. It cannot see structural conditions that only imaging or a physical exam would reveal.
Getting a more reliable reading
If you're using a camera-based posture check, a few habits improve consistency: same location, same lighting, same distance from the camera, minimal or fitted clothing around the shoulders/neck, and standing in a natural, relaxed position rather than "posing" for the photo. Repeating the scan under the same conditions each time makes week-to-week comparisons much more meaningful than any single reading in isolation.
Backed (our app) uses this same pose-estimation approach for its AI camera scan, paired with guided on-screen positioning to keep readings consistent, and turns the result into a personalized daily corrective exercise plan. It's a starting point for awareness, not a medical scan. Try Backed.
To see how this fits into a broader self-check, read our posture check guide, and for how different apps compare on this front, see our best posture app roundup.
Frequently asked questions
What angle do posture apps usually measure for forward head posture?
Most use the craniovertebral angle (CVA) - the angle between a horizontal shoulder line and a line to the ear. A smaller angle generally indicates more forward head posture, though exact reference ranges vary by source.
Does the app need a special camera or sensor?
No. Standard phone cameras are sufficient for pose-estimation-based posture checks; no depth sensor or special hardware is required, though consistent lighting and framing improve accuracy.
Is the AI actually seeing my skeleton?
No. It detects visible surface landmarks (like the ear and shoulder) in the image and calculates angles from those - it doesn't have information about bones, discs, or nerves underneath.
Can lighting or clothing really throw off the reading?
Yes, meaningfully. Poor lighting or loose clothing can obscure landmarks and shift the measured angle by several degrees, which is why consistent setup matters more than any single photo.
Is on-device processing more accurate than cloud processing?
Not necessarily - accuracy depends more on the model and image quality than where the calculation happens. On-device processing does typically keep your photo from being uploaded.
Should I trust a single scan result?
Treat one scan as a rough starting point, not a fixed number. Trends across multiple scans taken under similar conditions are more informative than any single reading.
Read next
Fix the cause, not just the reminder
Backed scans your posture and builds a personalised 5-minute-a-day plan to correct it. Free to start on iOS and Android.
Related guides
Posture Correction: The Complete Guide
The master guide - causes, timelines, exercises and every cluster.
Best Posture App
How Backed and the main posture apps honestly compare.
Neck Hump (Dowager’s Hump) Fix Guide
What it is, why it forms, and how to reverse it.
Rounded Shoulders Fix Guide
Causes, the fix timeline and the exercises that work.
Upper Cross Syndrome Guide
The muscle-imbalance pattern behind forward head and rounded shoulders.
How to Check Your Posture at Home
A phone-camera self-assessment you can do in five minutes.