How accurate are running form apps?
Running form apps are accurate enough to coach from for 2D angles measured from a well-filmed side view, such as landing reach, knee bend, torso lean, arm angle and cadence, but they cannot measure force, pronation or 3D rotation, and their results are only as good as the clip.
What they measure well
Pose estimation on a modern phone places joints within a few pixels on a clear, well-lit, side-on clip. For a runner filling a third of the frame, that is a small fraction of leg length. It is enough to tell:
- whether the ankle lands ahead of the hip, and roughly how far,
- whether the knee is straight or soft at contact,
- how far the torso leans,
- the elbow angle,
- whether the head sits forward,
- cadence, which is a simple count and is very reliable.
Those are the things a coach comments on. For those, a phone app and a coach with a slow-motion video usually agree.
What they measure badly or not at all
- Force. How hard you hit the ground needs a force plate. No camera measures it.
- Pronation. The foot rolling in or out is a small, fast rotation that a side view cannot see and a behind view sees only roughly.
- 3D rotation. Hip rotation, pelvis tilt in the walking direction and knee rotation are invisible or distorted from one camera.
- Left versus right subtleties. From the side you see one leg clearly and the other partly hidden.
- Why anything hurts. No app can diagnose.
The bigger risk: the clip
Most wrong results come from the video, not the model. A phone on the ground makes everyone lean back. A backlit runner becomes a silhouette. Two people in shot confuse the tracker. Portrait orientation gives two steps. A sprint gives sprint form.
The honest apps check the clip and refuse to score it when the runner is too small, too dark or out of frame. Caden does that. If an app scores every clip you throw at it with total confidence, be suspicious.
How to check a result yourself
- Look at the skeleton overlay on the freeze frames. Do the dots sit on your ankle, knee, hip, shoulder? If the knee dot floats mid-thigh, discount the knee numbers.
- Film twice, same setup, same pace. Scores should land within a few points of each other. If they swing 20 points, the clip is the problem.
- Compare the landing frame with your own eyes. You can see reach without a number.
How to check it with Caden
- Tap Run a new analysis, Side view, and follow the framing guide.
- If the clip is too dark, too far or too short, Caden says so before scoring. Refilm rather than accept a guess.
- On the results screen tap Breakpoints and turn on Skeleton. Check the joints on the foot strike frame.
- Each metric page says what a 2D camera can and cannot tell you about that measurement.
FAQ
Is a phone running form app as accurate as a gait lab?
For 2D angles from a side view, close enough to coach from. For force, pronation and 3D joint motion, no. A lab measures things a camera cannot.
Does slow motion make running form analysis more accurate?
A little for foot strike timing, not for angles. Normal 30 or 60 frames per second is enough. Filming and lighting matter far more than frame rate.
Why do two apps give different scores for the same video?
They pick different metrics, different ranges and different weightings. Compare trends within one app, not scores across apps.