Louis Rollet

Research / Time Series / Machine Learning

BioTemporal-HAR

Time-series research proposal for activity recognition and biometric identity verification.

Role
Author
When
Mar 2026 — May 2026
Status
submitted

BioTemporal-HAR studies smartphone and smartwatch inertial data for activity classification, closed-set subject identification, and identity verification. The proposal combines windowed signal processing, feature selection, compact CNN-LSTM sequence modeling, multi-sensor fusion, and Signal Detection Theory calibration to make false-alarm behavior explicit.

One signal, two questions

Smartphone and smartwatch inertial streams carry more than activity — they carry identity. BioTemporal-HAR asks whether one compact model can answer both what you are doing and whether you are you.

walkingrunningstanding

Method

  • Windowed signal processing — sliding windows over multi-sensor streams.
  • CNN-LSTM — compact sequence model for activity and identity heads.
  • Multi-sensor fusion — phone + watch inertial channels combined.
  • Signal Detection Theory — calibration that makes false-alarm behavior explicit.

The paper

Louis Rollet, Vincent Montero Fontaine. “BioTemporal-HAR: Human Activity Recognition and Biometric Identity Verification from Wearable Sensors.” Tsinghua University, Time Series Analysis, 2026. Under review on OpenReview.

OpenReview submission ↗
  • Human Activity Recognition
  • Time Series Classification
  • Behavioral Biometrics
  • Wearable Sensing
  • Multi-Sensor Fusion
  • Signal Detection Theory
  • CNN-LSTM
  • Identity Verification
  • Feature Selection

Project timeline

Tsinghua time-series research project exploring whether a single model can handle both activity recognition and biometric identification.

  1. Mar 2026 ~ Idea

    Research hypothesis defined

    Defined the hypothesis that a single model could perform both human activity recognition and biometric identification from time-series data.

  2. Apr 2026 ~ Development

    Modeling and experiments

    Worked on the research experiments and model comparison for the time-series analysis course.

  3. May 2026 ~ Release / Delivery

    Paper submission / publication draft

    Prepared the research-paper style project submission and public paper entry.

Built with

  • Python
  • Time-series ML
  • CNN-LSTM
  • Signal Detection Theory
  • Wearable sensing
  • Feature selection