RecoFit

Abstract

Although numerous devices exist to track and share exercise routines based on running and walking, these devices offer limited functionality for strength-training exercises. We introduce RecoFit, a system for automatically tracking repetitive exercises -such as weight training and calisthenics -via an arm-worn inertial sensor. Our goal is to provide real-time and post-workout feedback, with no user-specific training and no intervention during a workout. Toward this end, we address three challenges: (1) segmenting exercise from intermittent non-exercise periods, (2) recognizing which exercise is being performed, and (3) counting repetitions. We present cross-validation results on our training data and results from a study assessing the final system, totaling 114 participants over 146 sessions. We achieve precision and recall greater than 95% in identifying exercise periods, recognition of 99%, 98%, and 96% on circuits of 4, 7, and 13 exercises respectively, and counting that is accurate to ±1 repetition 93% of the time. These results suggest that our approach enables a new category of fitness tracking devices.

Asynchronous federated learning system for human–robot touch interaction

“…As the reader might have perceived, the system bears some resemblance to human activity recognition (HAR) systems (Kumari et al, 2021;Morris et al, 2014;Muaaz et al, 2021) (mainly in the realtime signal processing phase) since the information extracted from a microphone could also be extracted from an HAR device such as an accelerometer. The main difference between this kind of system and the one described in this section is that HAR systems, as their name implies, gravitate around a person doing an activity, while in our case, the central reference point is a robot.…”

Sense and Accessibility

“…New technologies such as 3D printing and other fabrication options enable embedding of sensors in a variety of materials and objects [63], which could have implications for accessibility (e.g., smart prosthetics [33]). Wearable technologies embed sensors directly into clothing or accessories, often with the goal of augmenting cognition (e.g., Google Glass's heads-up display) [58,62] or health tracking (e.g., many smart watch systems [17,42], some of which researchers have explored making more accessible [13,16]); Carrington et al have also explored expanding the concept of wearables to augment mobility aids [12]. Improved voice-based sensors are increasingly available as smart-speakers or phone-based virtual assistants (a category of sensor that may be particularly of interest to people with disabilities [10,46,59]).…”

Recognition and Repetition Counting for Complex Physical Exercises with Deep Learning

“…Human activity recognition (HAR) is an important sub-field of human-computer interaction (HCI), with a rich body of work. HAR covers a broad range of tasks, including distinguishing activity from non-activity [5,6], activity classification [7] and repetition counting [5,6,8,9,10]. These tasks are interesting by themselves from a research perspective, but also have a wide range of potential real world applications, especially in the fields of healthcare and personal fitness.…”

“…Similar to [9], they also used auto-correlation for repetition counting. Morris et al [6] built and improved on the aforementioned works. They used an arm-mounted inertial measurement unit and achieved up to 99% recognition accuracy across 14 gym exercises (plus walking and running).…”

“…We compared using a single smartwatch worn on the wrist to using an additional smartwatch worn on the ankle. With the exception of [6], most previous works mainly focused on single-joint exercises (e.g., bicep curl) or other, mostly stationary, exercises (e.g., bench press), while we chose ten full body exercises that included complex movements, such as burpees which consist of multiple “sub-exercises”. The exercises we chose vary greatly in execution time and form across athletes, making this a challenging task.…”