IMU Gait Analysis, Walking vs. Running

A wearable dual-IMU setup that captures and compares the biomechanics of walking and running from foot-mounted motion sensors.

A focused study on wearable gait analysis. I used two WitMotion WT901WIFI IMUs, one on the dorsum of each foot, to record and compare the biomechanics of walking and running in real time. The project covers the full signal pipeline: synchronized motion capture streamed over Wi-Fi, cleaning and zero-lag filtering, gait-cycle segmentation from heel-strike peaks, and statistical feature extraction that quantifies how the two gaits differ. Published as Human Motion Gait Analysis Using IMU Sensors and supported by USDA-NIFA Grant 2023-77040-41262.

One WitMotion WT901WIFI mounted on the dorsum of each foot, capturing 3-axis acceleration and angular velocity during treadmill trials.

At a glance

   
Role Lead author; data collection, signal processing, and analysis
Sensors 2× WitMotion WT901WIFI (MPU9250 9-axis IMU), foot-dorsum mounted
Signals 3-axis linear acceleration + 3-axis angular velocity, 10 Hz, over Wi-Fi
Capture Treadmill (Precor TRM 835), separate walking and running trials
Processing Python/pandas cleaning, 4th-order Butterworth low-pass (4 Hz, zero-lag)
Analysis Heel-strike peak detection, gait-cycle segmentation, statistical features
Funding USDA-NIFA Grant 2023-77040-41262

Why it matters

Clinical gait labs using multi-camera systems like Vicon and OptiTrack are accurate but expensive, and confined to a setting that doesn’t reflect how people actually move. Foot-mounted IMUs are cheap, portable, and capture both the stance and swing phases continuously, which makes gait screening practical outside the lab, in rehabilitation, sports, and everyday monitoring.

Setup and method

The dorsum was chosen as the mount point because its flexibility gives a clean, detailed picture of foot motion through each phase. Sensors on both feet allow a direct left/right comparison. Raw data was streamed over Wi-Fi into the WitMotion app, exported, then cleaned in pandas down to the acceleration and gyroscope channels. After a 4th-order Butterworth low-pass filter (4 Hz cutoff, zero-lag), I computed the resultant acceleration and angular velocity magnitudes to remove orientation dependence, then detected heel strikes as peaks in the acceleration signal to segment individual gait cycles.

Left: the stance and swing phases of a normal gait cycle, the framework for interpreting the signals. Right: the WT901WIFI mounted on an adjustable Velcro strap.

Walking vs. running

Segmenting and comparing cycles surfaces a clear biomechanical signature. Walking is slower and more deliberate: cycles of roughly 1–1.2 s, with acceleration peaks near 5 m/s² and angular velocity peaks near 500 °/s, and well-defined troughs between phases. Running compresses everything: cycles of 0.6–0.8 s, sharper peaks reaching ~6 m/s² and ~600 °/s, higher variability, and a right-skewed distribution reflecting explosive, intermittent bursts of force.

Resultant linear acceleration for both feet during walking (left) and running (right). Running shows shorter intervals between heel strikes and higher peak impacts.

What this project demonstrates

  • Wearable sensing. Multi-sensor, dual-foot motion capture with wireless data streaming.
  • Biomechanical signal processing. Zero-lag Butterworth filtering and orientation-independent magnitude derivation.
  • Gait-cycle analysis. Heel-strike event detection, cycle segmentation, and statistical feature extraction.
  • Comparative kinematics. Quantifying how walking and running differ in cycle duration, impact, and rotational dynamics.