Skip to main content

mmWave Radar Sensor Data Collection Session 1: Gait Recognition

1. Overview​

Project Name: mmWave Radar Gait Recognition (PoC)
Data Collection Session: 1st Session
Date: Fri 23 Jan
Location: Physics Building G12 Mott Lecture Theatre

2. Purpose and Scope​

Purpose: To collect high quality micro-Doppler radar signatures of human walking patterns (gait). This data will be used to train and validate a 1D Convolutional Neural Network (CNN) for person identification. The goal is to distinguish between specific individuals based solely on the unique velocity patterns of their limb movements.

Scope:

  • Participants: 2 Subjects (Alina, Michal).
  • Activity: Walking perpendicular to the radar (Left -> Right and Right -> Left) at around 3 meters away from the sensor.
  • Duration: Data is segmented into around 5 minute clips per json file.
  • Target Volume: 1 clip per participant.

3. Setup & Configuration​

Hardware Configuration​

  • Sensor Model: IWRL6432

  • Mounting:

  • Height: 1.0m – 1.2m (Chest/Waist height is optimal for gait; 1.5m is acceptable but lower captures legs better).

  • Orientation: Standard (Horizontal) is typically preferred for gait to maximize field of view.

  • Capture Zone:

  • Walk Path: A straight line marked on the floor, 3.0 meters from the radar.

  • Walk Width: Markers placed 2.5m to the left and 2.5m to the right of the center (5m total path).

Data Output​

  • Format: .json (JSON)

4. Participant Details​

Paricipant IDNameHeightRole
P0Alinaaround 170 cmTarget Class 0
P1Michalaround 180 cmTarget Class 1

5. Activity Log​

The data collection duration is measured in minutes, 5 per person.

Set IDParticipantTarget minutesStatus
Set 01Alina5[x]
Set 02Michal5[x]

6. Procedure (The 3-Second Loop)​

  1. Get Ready: Participant stands at the "Start Marker".
  2. Trigger: Operator starts recording.
  3. Action: Participant walks at a normal, comfortable pace along the 3m tape line.
  4. Cut: Recording stops after 5 minutes manually by operator.

7. Quality Control Checklist​

  • Gait Cycle Check: Does the spectrogram show at least 2 clear "flashes" (limb swings)?
  • Distance: Is the participant strictly following the 3m line? (Drifting closer changes the signal intensity).
  • Fatigue: If participant starts limping or walking lazily, take a break. The model needs "natural" gait.

8. Next Steps​

  • Preprocessing: Convert Micro-Doppler data from JSON to numpy.
  • Model training: Implement a script to transform the data into the shape of model input.