top of page
Co-Organisers:
I-RICE'26
4th International Research, Innovation, Creativity & Engineering Project Competition 2026
.png){kind=small}
Project ID:
ITCS04304
Wi-Fi 6 CSI-Based Fitness Movement Sensing
Project Title:
Category:
Information Technology/ Computer Science/ AI
Inventors:
ZOW CHENG YI, EE YUAN BING, NG YONG HAO, LIM XIN ZE
Institution/Company:
Southern University College
Invention Description/ Abstract:
With the growing popularity of home fitness, many people exercise without professional guidance, making it difficult to monitor exercise form and movement quality. This project proposes a Contactless Smart Fitness Monitoring System that uses Wi-Fi Channel State Information (CSI) to monitor human movement without wearable devices or cameras. The system provides a private, convenient and potentially low-cost approach for exercise monitoring in home and gym environments.
The system is designed for common exercises such as push-ups and squats. It provides the following functions:
Smart Repetition Counting and Rhythm Tracking: Detects repeated movement patterns, counts repetitions and monitors exercise rhythm.
Movement Form Assessment: Analyses changes in movement patterns to distinguish between selected correct and incorrect repetitions based on predefined movement characteristics.
Non-Intrusive Monitoring: Users do not need to wear sensors or position themselves in front of a camera. The system operates through wireless signals within the sensing environment.
The system consists mainly of a Wi-Fi transmitter and receiver. When a person moves within the wireless sensing area, body and limb movements cause changes in the propagation of Wi-Fi signals. These changes are captured through CSI data. Signal processing and classification methods are then used to extract movement-related features and identify selected exercise patterns.
Compared with conventional camera-based and wearable-based approaches, the proposed system explores a different sensing method based on wireless communication signals. It does not require visual recording and avoids attaching physical sensors to the user's body. The use of commercially available Wi-Fi hardware also provides potential for a low-cost implementation.
The system has potential applications in home fitness, gyms and basic rehabilitation monitoring, particularly where privacy, convenience and ease of use are important. Future development may include supporting more exercise types, improving detection accuracy, providing real-time feedback, and applying AI-assisted movement classification and movement quality assessment.
Invention Technical Description
This system uses Wi-Fi 6 Channel State Information (CSI) to capture how human movement alters wireless signal propagation. The setup consists of an Ai-M62-12F (TX) and an ESP32-C6 (RX), worn by the subject. CSI is derived from each subcarrier's preamble, giving per-subcarrier amplitude and phase.
The Ai-M62-12F runs as a software AP, transmitting continuously; the ESP32-C6 captures the resulting CSI. Wi-Fi 6's HE-LTF provides 256 subcarriers; as the subject moves, their body alters the propagation path, causing measurable CSI changes.
Research-grade CSI methods (e.g. Intel 5300 NIC) need a full PC runtime, impractical as a product. Pairing two low-cost Wi-Fi SoCs removes this dependency. More subcarriers give finer resolution and more information: legacy preambles yield ~56 subcarriers vs. 256 for HE-LTF. Since ESP32 cannot itself run as a Wi-Fi 6 soft AP (limited to legacy CSI in that role), we paired the Ai-M62-12F (as a Wi-Fi 6 soft AP) with an ESP32-C6 as the CSI-capturing station — giving high portability, low environmental dependency, and full 256-subcarrier CSI.
Captured CSI is denoised, then motion features are extracted to detect the target movement; for repetitive movements, detected events drive repetition counting. We have implemented and validated an ML event classifier (decision tree / SVM) distinguishing events from static periods via hand-crafted CSI-amplitude features.
The system is designed to be evaluated across movement types, distances, placements, and environments, using metrics such as detection accuracy, repetition-count accuracy, latency, and signal stability.
This project explores the information CSI offers for fitness-movement sensing. Body-induced electromagnetic attenuation carries information a camera cannot directly access — e.g., sensing unaffected by occlusion or lighting — though a single CSI link has far lower spatial resolution than a camera image; the two are complementary, not simply "more vs. less." The main limitation is that CSI datasets are far harder to build than image datasets.
This architecture lays groundwork for further work: an ML event classifier is already implemented; planned work includes richer signal processing, multi-movement classification, real-time feedback, movement-quality (form) assessment, and mobile/web integration. Environmental interference, position changes, movement variation, and other nearby moving objects may affect performance.
Demostration/ Presentation Video
Poster/ Broucher/ Invention Photo
Additional Documents
bottom of page