Semantic Scholar Open Access 2025

Pose Based Human Action Recognition for Accident Prevention in Industrial Environment

Aleksei Obshatko A. Bakhshiev Natalia Kosharich

Abstrak

Due to advancements in the fields of machine learning and computer vision, compliance monitoring systems for industrial safety by enterprise workers have become widely adopted in automating the solution of various classes of tasks. The focus of this work is on automating the monitoring of safety violations to prevent accidents. The specific task of this work is to monitor compliance with safety regulations when moving on stairs. An important direction in solving such tasks is the development of computer vision systems aimed at detecting human actions. This work proposes using a neural network for human pose estimation as a feature extraction network. A distinctive feature of the proposed solution is the implementation of a decision-making algorithm based on an LSTM network with a small number of parameters, and the use of optimized decisionmaking methods for operation on devices with limited computational power. All utilized neural networks were exported to TensorFlow Lite format to enhance performance when running on a central processing unit. The highest achieved accuracy was 98.2%, with a precision of 97.8% and a recall of 98.5% at a confidence threshold of 0.5.

Penulis (3)

A

Aleksei Obshatko

A

A. Bakhshiev

N

Natalia Kosharich

Format Sitasi

Obshatko, A., Bakhshiev, A., Kosharich, N. (2025). Pose Based Human Action Recognition for Accident Prevention in Industrial Environment. https://doi.org/10.1109/ICIEAM65163.2025.11028374

Akses Cepat

Informasi Jurnal
Tahun Terbit
2025
Bahasa
en
Sumber Database
Semantic Scholar
DOI
10.1109/ICIEAM65163.2025.11028374
Akses
Open Access ✓