Hang-Time HAR: A Benchmark Dataset for Basketball Activity Recognition using Wrist-Worn Inertial Sensors

TL;DR

This work presents a benchmark dataset for evaluating physical human activity recognition methods from wrist-worn sensors, for the specific setting of basketball training, drills, and games, and illustrates the dataset’s features in several time-series analyses.

Abstract

We present a benchmark dataset for evaluating physical human activity recognition methods from wrist-worn sensors, for the specific setting of basketball training, drills, and games. Basketball activities lend themselves well for measurement by wrist-worn inertial sensors, and systems that are able to detect such sport-relevant activities could be used in applications toward game analysis, guided training, and personal physical activity tracking. The dataset was recorded for two teams from separate countries (USA and Germany) with a total of 24 players who wore an inertial sensor on their wrist, during both repetitive basketball training sessions and full games. Particular features of this dataset include an inherent variance through cultural differences in game rules and styles as the data was recorded in two countries, as well as different sport skill levels, since the participants were heterogeneous in terms of prior basketball experience. We illustrate the dataset's features in several time-series analyses and report on a baseline classification performance study with two state-of-the-art deep learning architectures.

Figures (16)

• Figure S1: A scene and activities from the dataset: Offensive play of player 12 (yellow) and player 6 (red), see Table 6, with player 12 dribbling the ball (1), (2) and then passing (3) it to player 6. Player 6 then performs a layup (4). Video frames 1--4 and the performed activities are highlighted in the time-series below. The activity running is marked as yellow, layup as red, dribbling as mauve and pass is colored in blue.
• Figure S2: Our study design used 24 subjects with 13 subjects living in Germany and 11 subjects living in the United States of America. In each study, the players simultaneously performed the drills and game while the entire basketball court was monitored using two wide-angle cameras. After the study, the camera footage was used for detailed annotation of all activity-relevant data.
• Figure S3: Our custom smartphone app was used to synchronize all smartwatches' real-time clocks at the beginning of each recording through Bluetooth Low Energy (BLE) serial commands and start recording simultaneously. After the app is started, it first scans for all available Bangle.js smartwatches. After that, the user has the option of either starting all devices simultaneously or individually. Before the smartwatches are started, the user is asked to enter the desired parameters (sampling rate, sensitivity, and start time). After pressing the start button, all smartwatches are started with the desired parameters.
• Figure S4: Illustration of the multi-tier labeling approach, depicting the inertial data of subject 05d8_eu (top), the ground truth locomotion Layer (middle), and the ground truth basketball layer (bottom).
• Figure S5: Exemplar time-series data for the included activities. The examples shown for the periodic activities sitting, standing, walking, running, and dribbling contain 1200 samples (approx. 24 s). To better represent the complex activities shot and layup as well as the micro-activities pass and rebound. Jumps are marked in classes where the activity occurs. Such short periods were summarized in the activity jumping.