Activity trackers don't just capture data related to a specific activity—they’re actually capturing every move and vibration you make. So the effectiveness of one of these devices hinges on how well it can separate the noise from data. When are you running, and when are you riding the subway?
As the field of activity trackers gets more crowded, some device makers are boasting sport-specific algorithms that claim to do a better job with the noise-to-data ratio than more generalized trackers. Shot Stats Challenger is one example, recently launched on Kickstarter, that tracks the moves of tennis players and attempts to bring the data to life by combining it with video footage and other contextual information.
“The raw data from a bunch of sensors—in any application—is easy enough to get,” says Shot Stats cofounder Sergey Feingold. “The harder part is calibrating the sensors and making sense of all the data. Take racquet head speed for example—you want the speed of the racquet when it makes contact with the ball. However, there are no sensors that can be put onboard to measure racket speed directly.”
The Challenger attaches to the tennis racquet and can then gives insight into things like speed, pitch, technique, and other things related to how a person is playing tennis. It's essentially measuring two metrics: acceleration, and time/point of impact. Working backwards they can narrow down the parts of the motion and identify which part of the user's swing corresponds to which data readings—a huge challenge for anyone building a sport-specific tracking device.
“Even a simple (at first glance) device like Challenger shares a lot of the obstacles of considerably more sophisticated systems like autonomous cars and robots,” Feingold explains. “The scale of the problems is different, but the underlying concepts are similar. How do you effectively interpret data? How do you effectively find patterns based on a huge and chaotic stream of raw data? How do you make decisions based on those patterns in an intelligent and repeatable way?”
This isn’t just about tennis, though. Feingold says that the work his company is doing as it relates to a sports gadget is relevant across the board on many different levels. Misfit Wearables is another device tracking company which had a similar crowdfunding start to Shot Stats. Matthew Diamond, medical lead for Misfit, explains the first rubric for a good activity tracker is something like a Turing test.
“When you’re trying to decipher a large stream of raw data from a monitoring device with the goal of generating insights, the first level of analysis involves trying to develop algorithms that begin to provide what a good human coach could impart,” he says.
Developing an algorithm that can supplant a coach requires three steps, says Diamond:
- A stream of familiar data—namely, video recording—that is collected in a parallel fashion to the stream of data of the type you’re trying to decipher.
- Expert content knowledge that can generate insight from that familiar data so you know what insights you’re looking to derive from the undeciphered data.
- Advanced data science skills for formatting, retrieving, processing, and visualizing large amounts of heterogeneous data.
Similarly, Moov announced a yet to be released hardware device earlier this year that claims to be more of a personal trainer than activity tracker. Even more intriguing is that it supposedly isn’t limited to one sport, but a wide range of physical activities.
The idea is that no one really wants to get a simple step count, bur rather they actually want to be told how those steps are making a difference. Part of the simplistic nature of the current crop of activity trackers might be that these data interpretation algorithms are still so nascent. Fortunately for third-party developers, a lot of the companies facing these initial data collection hurdles are also providing open API access, which should speed up the improvement of these sorts of algorithms.
[Image: Flickr user Magdalena Roeseler]