HEAR: Human Action Recognition via Neural Networks on Homomorphically Encrypted Data

04/19/2021
by   Miran Kim, et al.
0

Remote monitoring to support "aging in place" is an active area of research. Advanced computer vision technology based on deep learning can provide near real-time home monitoring to detect falling and symptoms related to seizure, and stroke. Affordable webcams, together with cloud computing services (to run machine learning algorithms), can potentially bring significant social and health benefits. However, it has not been deployed in practice because of privacy and security concerns. People may feel uncomfortable sending their videos of daily activities (with potentially sensitive private information) to a computing service provider (e.g., on a commercial cloud). In this paper, we propose a novel strategy to resolve this dilemma by applying fully homomorphic encryption (FHE) to an alternative representation of human actions (i.e., skeleton joints), which guarantees information confidentiality while retaining high-performance action detection at a low cost. We design an FHE-friendly neural network for action recognition and present a secure neural network evaluation strategy to achieve near real-time action detection. Our framework for private inference achieves an 87.99 sensitivity and 99.14 seconds on real-world datasets. Our evaluation shows that our elaborated and fine-tuned method reduces the inference latency by 23.81 straightforward implementation.

READ FULL TEXT
research
07/21/2022

Real-Time Elderly Monitoring for Senior Safety by Lightweight Human Action Recognition

With an increasing number of elders living alone, care-giving from a dis...
research
07/08/2019

A Deep Learning Approach for Real-Time 3D Human Action Recognition from Skeletal Data

We present a new deep learning approach for real-time 3D human action re...
research
06/13/2020

3DFCNN: Real-Time Action Recognition using 3D Deep Neural Networks with Raw Depth Information

Human actions recognition is a fundamental task in artificial vision, th...
research
06/06/2018

Action4D: Real-time Action Recognition in the Crowd and Clutter

Recognizing every person's action in a crowded and cluttered environment...
research
06/08/2023

Human Action Recognition in Egocentric Perspective Using 2D Object and Hands Pose

Egocentric action recognition is essential for healthcare and assistive ...
research
12/27/2018

Low Latency Privacy Preserving Inference

When applying machine learning to sensitive data, one has to balance bet...
research
04/14/2023

Skeleton-based action analysis for ADHD diagnosis

Attention Deficit Hyperactivity Disorder (ADHD) is a common neurobehavio...

Please sign up or login with your details

Forgot password? Click here to reset