Workout Type Recognition and Repetition Counting with CNNs from 3D Acceleration Sensed on the Chest

Abstract

Sports and workout activities have become important parts of modern life. Nowadays, many people track characteristics about their sport activities with their mobile devices, which feature inertial measurement unit (IMU) sensors. In this paper we present a methodology to detect and recognize workout, as well as to count repetitions done in a recognized type of workout, from a single 3D accelerometer worn at the chest. We consider four different types of workout (pushups, situps, squats and jumping jacks). Our technical approach to workout type recognition and repetition counting is based on machine learning with a convolutional neural network. Our evaluation utilizes data of 10 subjects, which wear a Movesense sensors on their chest during their workout. We thereby find that workouts are recognized correctly on average 89.9% of the time, and the workout repetition counting yields an average detection accuracy of 97.9% over all types of workout.