Published in 2022 IEEE 2nd Mysore Sub Section International Conference (MysuruCon)

Link: https://ieeexplore.ieee.org/document/9972571

This paper presents a Wearable Fitness Monitor (WFM): a chest-mounted embedded device designed to measure and count walking, running, shuttle runs, push-ups, and sit-ups with reasonably high accuracy. The work is motivated by widespread physical inactivity despite the growing popularity of consumer fitness trackers, and it focuses on exercise-specific measurement rather than generic activity tracking. The system uses an IMU because inertial sensing can be integrated into compact wearable hardware and supports real-time motion analysis; chest placement is selected to obtain stable trunk-motion signals across both locomotion and calisthenic exercises. The authors position the device as useful not only for personal fitness but also for structured athletic, military, or organizational fitness assessment, where consistent exercise form and reliable repetition counting matter. The underlying technical approach—extracting meaningful movement features from wearable IMU data—aligns with broader evidence that chest-mounted inertial sensing can support high-accuracy activity recognition.

Architecturally, the WFM is a self-contained embedded system built around a 32-bit ARM Cortex-M0+ microcontroller, IMU sensor sampled at 20 Hz, micro-SD storage, LEDs, push buttons, vibration motor, battery-voltage measurement, and USB mass-storage connectivity. Firmware simultaneously acquires motion data, executes exercise-detection algorithms, records results, and continuously performs built-in tests for battery status, IMU identity/calibration and I2C health, SD-card presence and read/write capability, and correct chest-harness attachment. This is a strong system-engineering feature because it treats reliability, fault logging, user feedback, and data integrity as part of the wearable’s functional design rather than as afterthoughts. A desktop application complements the device by providing authenticated subject-data management, exercise-data visualization, and historical-result review. The chest harness also includes a removal/orientation detection mechanism through spring-loaded electrical contacts, helping an instructor identify invalid use or removal during testing.

The signal-processing strategy is deliberately lightweight and suitable for an embedded Cortex-M-class target rather than relying on computationally intensive machine learning. For walking and running, the WFM filters y-axis acceleration using a 21st-order FIR low-pass filter with a 4 Hz cutoff, then detects steps using amplitude thresholds, minimum peak separation, and continuous buffering to reduce false positives, missed peaks, and boundary-related undercounting. Shuttle-run detection additionally uses yaw-angle change to identify U-turns, with angle representation adjusted to avoid the discontinuity at the 0-degree/360-degree transition. For push-ups and sit-ups, the system estimates Euler angles through a complementary filter and uses the pitch angle as the principal feature: an 80-sample, four-second moving window is cross-correlated against an exercise template, then validated through correlation, mean, variance, and initial/final slope checks. This multi-stage statistical validation is important because it rejects incomplete repetitions and incorrect body movement rather than merely detecting periodic motion. Across 25 male participants with BMI 21–25, reported average accuracy was 97% for walking, 95% for running, 96% for push-ups, and 97% for sit-ups, with push-up and sit-up counts exhibiting up to ±3 repetitions of variability under the commanded two-second movement cadence. The results support the paper’s claim that deterministic IMU algorithms can provide useful real-time exercise monitoring, while its proposed future upgrades—GPS, machine-learning classification, broader protocols, encryption, and medical/security compliance—would improve scalability and deployment readiness.