Last week I wrote about the next frontier in baseball data: biomechanics. Using an open-source software package called OpenPose, I used bog-standard broadcast clips to get the coordinates of Mike Trout’s limbs as he swung (first a golf club, then a baseball bat). The possibilities of biomechanics data are limitless, stretching from pitch tell-hunting to prospect analysis to injury detection, and well beyond.
But digitizing video clips is only half the problem. The camera (and thus the computer) sees images only in terms of pixels: the tiny dots of color which together compose the visual we see. We don’t want to do biomechanical analysis on pixels though–it doesn’t make sense to say that Trout’s rotational velocity is 5000 pixels/second. So we need to turn pixels into the common units we think about: feet and inches, specifically. With a little help from Statcast, we can go a long way towards that goal.
Images, as encoded by video cameras, consist of a sequence of colors. The resolution of an image–say, 4K or 1920×1080, refers to how many colored dots make it up. Show the colors next to each other and your brain stitches recognizable objects out of them.