Camera position, frame rate, three visible features and three invisible ones: how to film and read your own push — and exactly where video analysis stops.
When you film your push, you look for exactly one feature first: the angle of attack. In an analysis of 305 ground contacts from 28 male elite bob athletes it was the strongest performance predictor across all push phases, and a random forest model explained 67, 60 and 60 per cent of the acceleration differences in start acceleration, acceleration phase and high-speed phase with it [1]. No other feature a camera can see is so tightly coupled to performance.
That is the good news for anyone without a biomechanics lab: the strongest known predictor is an angle — and angles can be filmed.
The bad news arrives immediately after. The strongest single correlation with push-start performance is Pmax at around 0.80 [4] — the maximum mechanical power from your sprint force-velocity profile. Pmax appears on no video. Video analysis does not replace performance diagnostics; it complements them exactly where diagnostics are blind: movement form.
And so you do not over-read the analysis: push time and finish time correlate moderately (around 0.48 men, 0.63 women), heat-to-heat changes not at all [5]. A single video is a snapshot. Twenty videos are a development.
The angle of attack lives and dies by perspective. A camera rotated by ten degrees produces an angle change you will read as progress. Mark the tripod point on the ground, note height and distance, and use exactly those values in every session. Without a fixed perspective the time series is worthless.
The research separates start acceleration (first ground contacts), acceleration phase at 10 metres and high-speed phase at 30 metres [2]. A panning phone delivers three distorted perspectives in one clip. Better: one fixed camera per zone, or the same zone across several runs.
First the angle of attack at defined contacts, in degrees, with a tolerance band [1]. Second the hip extension of the stance leg at ground contact — together with the swing leg's hip flexion velocity it carries forward acceleration [2]. Third the rhythm of contact and flight time, which you can count directly at a high frame rate. Three variables measured cleanly beat ten estimated.
In the push, bob athletes generate more mechanical energy at the hip and less at the ankle than in unresisted sprinting [1], and propulsion is hip-extensor dominant while the ankle contribution only grows with rising speed and the knee contribution falls [2]. On video that means: an athlete who looks like a 100-metre sprinter during start acceleration is probably working the wrong joint.
Three things you do not see. Pmax — the strongest single correlation with push performance at around 0.80, and with it V0, Vopt and relative Pmax [4]. The absolute system acceleration, as long as you have no calibrated markers and times [2]. And the surface effect: on ice, step frequency is higher and stance, swing and flight times shorter than on land, most clearly in the start phase [3] — that work is a preprint, but the consequence is unambiguous: a dry video is not an ice video.
Same camera, same markers, same three features, at least every two weeks. Only across twenty sessions do you see whether your angle of attack is drifting systematically or merely fluctuating — and given the weak heat-to-heat relationships [5] that is the only analysis that holds.
Video analysis tempts you into extra attempts: "one more, that one was messy". That is exactly how the typical push injuries to hamstrings and adductors arise. Fix the number of attempts before the session and film what happens — not until the picture looks nice.
Loading into the moving sled is the most delicate moment of the push and no place for an improvised camera position. Tripod, cables and people belong outside the lane and outside the run-out.
This guide explains analysis methodology, it does not replace coaching. Pushing, loading and every contact with sled or ice channel run exclusively through club, federation and qualified coaches. And self-correction from video has a limit: what you see is a symptom — your coaching team names the cause.
Film two runs every session: one fresh, one as the last of the unit — and compare only those two. The fresh run shows your technical optimum; the last shows what remains of your movement under fatigue. That gap is the most interesting number of your season, because in competition across four heats over two days it is not your best push that counts but your worst. If your angle of attack deviates clearly in the tired run [1], you do not have a technique problem but a capacity problem — and from tomorrow you train something else.
Above all the angle of attack: across 305 ground contacts from 28 elite athletes it was the strongest performance predictor over all push phases, and a model explained 67, 60 and 60 per cent of the acceleration differences with it [1]. Added to that is hip-extensor dominant propulsion with high hip extension velocity of the stance leg at ground contact [2].
Perpendicular to the run, fixed tripod position with documented height and distance, at least 120 frames per second and markers at 0, 10 and 30 metres — matching the phase logic of the research [2]. Repeatability is what matters: a changed perspective produces angle changes that look like progress.
Your power. Pmax is the strongest single correlation with push-start performance at around 0.80, complemented by V0, Vopt and relative Pmax [4] — none of it visible. Absolute system acceleration also needs calibrated markers and times [2]. Video analysis complements performance diagnostics, it does not replace them.
No. On ice, step frequency is higher and stride, stance, swing and flight times shorter than on land, with a smaller vertical centre-of-mass excursion — most clearly in the start phase [3]. Keep separate folders and separate time series. That study is a preprint but practically usable.
Regularly and identically, at least every two weeks with the same setup. Single runs say little: push time and finish time correlate only moderately, and heat-to-heat changes did not correlate with changes in finish time at all [5]. What counts is the trend across twenty sessions, not the best clip.
Women in Bobsleigh: What the Data Says Differently — and Where the Evidence Stops