HAR Demo

HAR

Human Activity Recognition · CIS4930

idle

WAITING

—

no tracker active

Smartphone-Based Human Activity Recognition Using Inertial Sensors

Mateo Saettone Tomas Pastore Godoy Ethan Thompson

project details

course: CIS4930 — Wireless and Mobile Computing
instructor: Dr. Yili Ren
school: University of South Florida · Spring 2026
dataset: UCI HAR — 30 subjects, 6 activities, 50 Hz accel + gyro
model: 1D CNN · 90.5% test accuracy (Random Forest baseline 85.3%)
pipeline: DeviceMotion → 2.56 s window → ONNX Runtime Web → Supabase Realtime Broadcast
source: github.com/MateoSaettone/Har-Demo

debug

body_acc source:—

event rate:—

body_acc (g):—

total_acc (g):—

body_gyro (rad/s):—

probs:

—