How to Film Your Own Running Gait at Home

Most gait videos that runners upload to an analysis tool are unusable, and the reason is almost never the runner’s form. It’s 30 frames per second, a phone held at chest height, baggy shorts, and a camera placed at an angle nobody can measure from. Fix those four things and a £0 setup in your garden or garage will produce footage good enough to track pelvic drop, footstrike position and cadence across a whole marathon block.

This page covers the filming only: the angles, the frame rates, the lighting, the exact protocol, and how to check the files before you put your kit away. If you want the wider picture of what the footage is actually for and which metrics survive contact with reality, the pillar on AI gait and form analysis from phone video covers that side.

Why frame rate is the single decision that matters

Ground contact time for a recreational runner at 5:00/km sits somewhere around 220 to 280 milliseconds. Work out how many frames of stance that gives you:

Peak rearfoot eversion typically happens 70 to 110 ms after initial contact, inside a window narrower than two frames at 30 fps. You will not catch it. You’ll catch something near it, and whatever tool you’re using will report a number with two decimal places that’s really a coin flip.

So: 240 fps if your phone offers it, 120 fps as the floor. On iPhone that means Slo-mo at 1080p, set under Settings > Camera > Record Slo-mo (every iPhone from the 11 onwards does 1080p at 240). Pixel 9 and Galaxy S24 both do 1080p at 240 in their slow-motion modes. Samsung’s 960 fps “Super Slow-mo” is not the one you want, because it drops to 720p and only captures a fraction of a second. A GoPro Hero 12 at 1080p240 is fine too and the fixed wide lens makes repeat setups easier.

Resolution is the thing you trade away happily. 1080p at 240 beats 4K at 30 for this job every single time.

The three angles, with actual distances and heights

Film all three if you can, and film them separately rather than trying to catch a diagonal that serves two purposes.

Rear view. Lens at knee height, roughly 0.4 to 0.6 m off the ground, on the exact midline of the treadmill belt or running lane. Stand 5 to 8 m back. This is the view for rearfoot motion, crossover and pelvic drop, and it’s the one people ruin by filming from standing height, which makes every heel look neutral.

Side view. Lens at hip height, 0.9 to 1.1 m, placed square to the direction of travel so the lens axis is perpendicular to your path. Sit it 3 to 4 m to the side. Off-perpendicular is the killer here: a camera rotated 15 degrees off square will change your measured knee flexion angle by several degrees, and since you’ll re-film in six weeks from a slightly different spot, that error becomes a fake trend.

Front view. Hip height again, 5 to 8 m ahead, on the midline. Useful for knee valgus and arm crossover, less critical than the other two.

No tripod is fine. A knee-height rear shot works from a stack of books, an upturned crate or a foam roller box. For side and front, a Joby GorillaPod at about £25 or a phone wedged into a gym bag on a chair will hold still enough. What you must not do is hand-hold, because any pan introduces apparent joint movement that isn’t there.

Turn off digital stabilisation if your app lets you. Action mode on an iPhone crops and warps the frame to keep the horizon level, which shifts pixel positions between frames and quietly corrupts any angle you measure later. Blackmagic Camera (free on iOS) and Filmic Pro both expose stabilisation and shutter controls that the stock app hides.

Light, because 240 fps eats it

At 240 fps each frame gets at most 1/240 s of exposure. That’s four stops less light than a normal 1/60 s video frame, and your phone will compensate by pushing ISO into the range where limbs turn into smears.

Film outdoors, in open shade or bright overcast, between roughly two hours after sunrise and two hours before sunset. Direct low sun from behind the camera is second best because it throws long shadows that confuse edge detection. A garage or spare room needs serious light: two 100 W LED work lamps at 10,000 lumens each, placed either side and slightly ahead of the runner, is about the minimum for clean 240 fps indoors. One ceiling bulb will not do it, whatever it looks like to your eye.

Check the footage for blur by scrubbing to a frame at mid-swing, where the foot is moving fastest. If the ankle is a ghost, add light or drop to 120 fps.

Make your body readable

Wear the tightest thing you own. Split shorts, tights, or a fitted tri suit. Baggy 7-inch shorts hide exactly the pelvic and femoral movement you’re filming for, and there is no software fix.

Mark landmarks with 15 mm coloured electrical tape or a white grease pencil, in a colour that contrasts with both your skin and your kit:

• Greater trochanter (side of hip), lateral femoral epicondyle (side of knee), lateral malleolus (outside ankle bone) for the side view
• Two dots on the vertical midline of each heel counter, 5 cm apart, plus one on the mid-calf for the rear view
• One on each ASIS (front hip bones) for the front view

For rear view, avoid black shoes against dark tarmac. Chalk or white tape on the heel buys you a lot.

A 12-minute protocol you can repeat

Warm up first, properly, for at least 10 minutes. Cold gait is not your gait.

1. Mark out a capture zone 8 m long on a flat stretch with at least 20 m of run-in either side. At 5:00/km you cross 8 m in 2.4 s, which is about four full gait cycles per pass.
2. Set the rear camera. Do three passes at easy pace, roughly 6:00/km, then three at 5K effort. Jog back between each.
3. Move to side view. Same six passes.
4. Move to front view. Three passes at easy pace is enough.
5. Then the one most people skip: go and run 40 minutes, come back, and film three more side-view passes at easy pace without any extra warm-up.

That last block is the valuable one if you’re auditing an AI plan. A runner who films fresh at 176 spm with 4 degrees of pelvic drop, and films at minute 45 of a long run at 166 spm with 9 degrees, has found something a Runna or Garmin Coach progression from 40 to 52 km/week will interact with. Fresh footage tells you almost nothing about whether your volume jump is safe.

On a treadmill, log the belt speed in the filename and keep it identical on every re-film. Treadmill gait differs from overground gait, which is fine, as long as you always compare like with like.

Verify the files before you stop

Do this at the trackside, not three days later.

$ ffprobe -v error -select_streams v:0 \
    -show_entries stream=width,height,r_frame_rate,avg_frame_rate,nb_frames \
    -of default=noprint_wrappers=1 IMG_4821.MOV

width=1920
height=1080
r_frame_rate=240/1
avg_frame_rate=239.76/1
nb_frames=4795

4795 frames at 240 fps is 19.98 s of real time, which matches the pass you filmed. If avg_frame_rate comes back as 30 or 29.97, your phone recorded normal video and rendered the slow motion on playback, and the real temporal resolution is gone.

A quick manual cadence check catches most remaining problems. Note the frame number of your first right footstrike and your tenth. If those are frames 312 and 1790, that’s 1478 frames for 9 right strides, so 6.16 s, 0.684 s per stride, 175 spm. Compare to what your Forerunner or Coros logged for that segment. Within 2 spm means your timing is trustworthy.

Getting the footage into a tool that measures things

Expect roughly 300 MB per minute of real-time 1080p240 HEVC, so a nine-pass session is about 1 GB. Kinovea (free, Windows) is still the most capable measuring tool, and it wants constant frame rate H.264, so convert:

ffmpeg -i IMG_4821.MOV -c:v libx264 -crf 18 \
  -vsync cfr -r 240 -an 2026-09-30_rear_easy-600.mp4

In Kinovea, set the capture frame rate to 240 under video properties so its timing readouts are real. Hudl Technique and OnForm handle side-by-side comparison against an older clip well, which is what you want when you re-film. Ochy will take a single side-view phone clip and return metrics automatically. Coach’s Eye, which older guides still recommend, was discontinued in 2021.

Name files with date, angle and pace, every time: 2026-09-30_side_5k-440.mp4. Six months of unlabelled IMG_ files is a dataset you can’t use.

What usually goes wrong

Book the next session into your calendar now, for the week before your plan’s biggest volume increase. That footage, at 240 fps, from a chalk-marked camera position, at a pace you wrote down, is the baseline everything after it gets compared against.