For The Wearable Revolution To Take Off, Accuracy Must Improve

Credit: ryoichi tanaka via Flickr

Credit: ryoichi tanaka via Flickr


Although many new wearable technologies bring functions that used to require bulky machinery to small devices, many products struggle in accurately tracking motion. The web is littered with side-by-side tests of accuracy of similar form factor devices, with results that call into question the entire category at times. Yet, as the technology and market for wearable devices develop, we think consumers will start to care about accuracy more and more, even though today, directionally accurate seems to trump actual accuracy in most devices for the consumer markets.  But how accurate will future devices need to be and how will device manufacturers and developers create more accurate tracking systems? We’ve often wondered about the challenges faced by the companies delivering accurate data in pleasing form factors, especially in covering CheckLight and X2 (for impact monitoring) and ‘PSM’ systems by Zephyr. In this post, we turn to a technical analysis to look inside accuracy, to understand how tracking is done and why it is difficult. We find that balancing data downloading, processing, and synthesis while managing battery life makes accurately tracking motion a difficult balancing act. Not to mention delivering it in a consumer-pleasing form factor.

The web is littered with side-by-side tests of accuracy of similar form factor devices, with results that call into question the entire category at times. Yet, as the technology and market for wearable devices develop, we think consumers will start to care about accuracy more and more, even though today, directionally accurate seems to trump actual accuracy in most devices for the consumer markets.

The first task is getting accurate data from sensors. To get data from sensors, most devices use Bluetooth wireless connections to mobile devices or USB connections to computers. Where devices vary is the number and quality of sensors they include. While more sensors can lead to more accurate data, it also often leads to more expensive devices or bulkier devices with poor battery life. The following table shows the number of sensors in several popular wrist wearable devices.

Table 1: Sensors, Price, and Battery Life of Selected Popular Wrist Wearable Devices

ProductBasisNike+ FuelBandUp By Jawbone (2nd gen)Flex By Fitbit
Number of Sensors4211
Types of SensorsOptical Blood Flow, Accelerometer, Perspiration Monitor, Skin TemperatureAccelerometer, Ambient LightAccelerometerAccelerometer
Quoted Usage Time3.5 days4 days10 days5 days

There are several challenges associated with the choice of sensors, though it is important to note that sensors are rapidly becoming commoditized. More advanced sensor will produce more complicated, difficult to process data. Putting in too many sensors will destroy the battery life. Cost can also be an issue. The balance here is therefore delicate. Wearable technology makers need to strike a balance between accuracy and feasibility in order to maximize the user experience. How do the devices in the table above compare?

For answers, we look to a review by Get Grok, which compares these devices (except Basis) with BodyMedia LINK, an arm worn device. The following graph from this review shows variance in step count between these devices over 28 days; note that Fitbit One and Jawbone Up were most accurate in a shorter test against a manual count at a walking pace. In addition, the review finds that devices vary greatly in terms of total step counts over 28 days—Nike+ recorded around 115,000, Jawbone UP and BodyMedia Link tracked around 150,000, and Fitbit One recorded more than 190,000 steps. Overall, the calories burned, sleep tracking, and active time results also vary greatly between devices; some devices consistently track more or less than others over the same period of time. The review also finds that Jawbone UP has the best software/app experience but also has the short battery life, which challenges the manufacturers’ quoted usage times in the table above. This variance between devices shows that accuracy is an issue in devices available today.

Table from Get Grok Review

Perhaps the best way to understand the various challenges faced by producers in choosing which sensors to include is to explore the process we would take to develop our own wearable activity tracking device for fitness and lifestyle applications. To begin, we need to determine the number, types, and accuracies of the sensors that shall go onto the device. Let’s cap the total number of sensors to be deployed at 3, for the sake of battery life and ease of data processing. To monitor the heart rate, we chose a new flexible ECG sensor similar to this wireless ECG sensor by IMEC. We use a galvanic skin response (GSR) system for our second sensors to measure the skin’s electrical conductance (linked to moisture level). GSR can monitor users’ mood and wellbeing by providing inside to the sympathetic nervous system (Read more about GSR). For last sensor, we would add a tri-axial accelerometer to monitor motion.

After transferring the data, activity tracking systems need to process it before then generating information and later making an interface to share the information with users. Much like a diamond needs to be cut before it can be polished, the data that we have collected needs to be filtered through various algorithms and techniques to get rid of unwanted noise. There are several proprietary engines to process data from sensors. This is a large element of some companies’ competitive advantage; we recently looked at Fullpower Technologies and MotionX, which might be an ideal acquisition target in part because of their advanced IP portfolio and know-how in this area. To reduce the need for huge storage capacities, some use algorithms resembling Kalman Filters. This efficient—but perhaps not the most accurate—method works in a two-step process. First, it predicts users’ future position based on their current position and the average trend (constant acceleration, moving forward, moving backwards etc.). Then, the algorithm updates the position of the user by taking a weighted average of the predicted position (which was calculated in the previous step) and the sensor data position (taken from the sensor). This refines datasets and provides a manageable, averaged continuum.

Form factor is another important technical aspect of activity tracking because it greatly influences data availability. While devices might be good at tracking one type of exercise, it is often difficult to detect all activities with one device because of limited motion and sensor data. For instance, a wrist wearable device might be inaccurate at tracking a bike ride or weightlifting session because its sensors may not effectively track the motion associated with these activities. A review Nike Fuelband Review in CNET notes that its difficult to use the device for weightlifting and yoga. This shows how devices can be limited by form factor, not to mention how devices can fail because of form factor. All one needs to do is see the troubles Jawbone’s initial UP product has had, largely due to its complicated design elements. By wanting to make the band flex at any point, new complexities were introduced into the electronics that drove failure rates to an unacceptable level. In other words, form and design elements rendered poor performance. (Editor’s Note: We look forward to the next generation of Jawbone Up, soon to be released. We hear that all of the electronic failures have been eliminated and a best of breed product is imminent and enroute, with Bluetooth syncing.)

Another difficultly is that the number of available form factors continues to increase. Today, devices ranging from wrist, ankle, and waist wearable devices to heads up displays are common, and new technologies are promising to bring sensors in flexible patches and fabrics. Accordingly, application and engine developers have now multiple challenges to support these devices, much more complicated than say, supporting multiple operating systems. Wearable device producers are challenged in selecting the best form factor to fulfill the needs of their device’s primary use. For new companies, form factor can also establish competition because consumers might have different perceptions of devices with similar functions but dissimilar form factors. These examples show how form factors can be a limitation and challenge in designing wearable devices.


Lastly, and perhaps most importantly, battery life presents a massive challenge to wearable device companies. Optimizing battery life requires tradeoffs with all the components mentioned above (number and quality of sensors, data transfer and processing, etc.). Recent developments in battery and energy capture technology have helped lessen the impact of this problem. Innovations in thermo electric generators along with other mobile charging methods, such as micro thermal energy harvesters from Micropelt GmbH, will likely improve the future of this space. Findings from academic research will likely impact wearable computing’s future as well; recent exciting updates include bendable, graphene micro-supercapacitors from UCLA and acquisition-cost aware continuous query processing from University of Hawai‘i (pulling data instead of pushing it).

Accurately tracking activity is challenging because of the tradeoffs required between sensors, processing data, battery life, supporting multiple form factors, and cost.  While accuracy for fitness and similar activity tracking is useful, today being directionally accurate has been as effective as actually accurate. However, it is clear that accuracy will be an important battlefield for both consumers and commercial uses in the future, especially if wearable devices will bleed into healthcare. New digital health systems and wearable medical devices must be accurate to be effective in transforming healthcare with accountability and proactive applications. In the near future, there will be many opportunities for innovation and profit in the categories above as accuracy becomes more important. We look forward to accuracy improving in the future and will continue to bring you updates about the latest updates about wearables from academic research, fresh startups, and established companies R&D.

, , , , , , ,

2 Responses to For The Wearable Revolution To Take Off, Accuracy Must Improve

  1. ForbesOste April 29, 2015 at 11:16 pm #

    Great article. I would like to use your graphic on accuracy trade-offs in wearables in my PhD dissertation on Wearables and Presence of Mind, with citation of course. Can you please tell me who I can contact for this? Thanks.

  2. admin May 4, 2015 at 9:00 pm #

    Feel free.

Leave a Reply