Related Experiment Video
Updated: Nov 27, 2025
06:25
Author Spotlight: Assessment of Visual Acuity in Central Vision Loss Through Motion-Based Peripheral Vision Testing
Published on: February 23, 2024
875
Action Recognition Using Single-Pixel Time-of-Flight Detection
Ikechukwu Ofodile1, Ahmed Helmi1, Albert Clapés2
1iCv Lab, Institute of Technology, University of Tartu, 50411 Tartu, Estonia.
Entropy (Basel, Switzerland)
|December 3, 2020
Summary
This study introduces a privacy-preserving action recognition method using scattered light pulses. Recurrent neural networks achieved over 96% accuracy in recognizing human actions without visual data.
Area of Science:
- Robotics
- Computer Vision
- Machine Learning
Background:
- Action recognition is crucial for robotic systems relying on visual data.
- Privacy concerns necessitate non-visual action recognition methods.
- Current methods often compromise user privacy.
Purpose of the Study:
- To propose a novel action recognition method preserving subject privacy.
- To utilize scattered light pulse data for action detection.
- To demonstrate the efficacy of machine learning for this task.
Main Methods:
- Recording temporal evolution of scattered light pulses using a single-pixel detector at 1 GHz.
- Embedding distance and shape information within light pulse traces.
- Applying recurrent neural networks (RNNs) for data analysis and action recognition.
Main Results:
- Successful action recognition demonstrated using the proposed method.
- Achieved an average accuracy of 96.47% for recognizing five distinct actions.
- The method effectively extracts action-related information from light pulse data.
Conclusions:
- The proposed light pulse-based method offers a privacy-preserving alternative for action recognition.
- Recurrent neural networks are effective for analyzing temporal light pulse data.
- This approach has significant potential for applications in privacy-sensitive robotic systems.
