.png){kind=icon}
.png){kind=icon}
Return
Pose Recognition System Using Bluetooth Low Energy and Machine Learning Techniques
.png){kind=icon}
OA
AI
DOI：10.1109/JIOT.2026.3660732.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This study addresses the challenge of robust human pose recognition in complex indoor environments by proposing a system that generalizes to unseen user locations using only angle-of-arrival (AoA) data from Bluetooth Low Energy (BLE) 5.1 devices. A comparative evaluation was performed, assessing a Random Forest model alongside three neural network architectures—convolutional neural network (CNN), long short-term memory (LSTM), and hybrid models—for identifying four physical states: standing, sitting, lying down, and walking. Data were collected in a real indoor environment using smart personal protective equipment (PPE) equipped with BLE transmitters placed on a helmet, upper and lower shirt regions, and a boot. The preprocessing pipeline included filtering, interpolation, feature extraction, and application of the fast Fourier transform (FFT). Models were evaluated under two scenarios: 1) known positions (used in training) and 2) unknown positions (never seen before). The best overall performance on unknown positions was achieved by the Attention-based BiLSTM–CNN, reaching an accuracy of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {A} = 94.89\%$ </tex-math></inline-formula> and a macroaveraged <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$F1$ </tex-math></inline-formula>-score of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {F}_{1} = 95.59\%$ </tex-math></inline-formula>, demonstrating strong generalization capabilities. In the scenario with known positions, the highest performance was obtained by the Random Forest, with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {A} = 99.95\%$ </tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {F}_{1} = 99.95\%$ </tex-math></inline-formula>, followed by the Deep CNN, with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {A} = 97.27\%$ </tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {F}_{1} = 97.28\%$ </tex-math></inline-formula>, closely followed by the Attention-based BiLSTM–CNN, which also showed robust classification accuracy across all posture classes. The analysis confirmed that using four wearable devices led to the highest performance, and sensor combinations significantly affected accuracy. The study also provides a publicly available dataset to support reproducibility and benchmarking.
Keywords:
Angle of arrival (AoA)
Bluetooth Low Energy (BLE)
neural networks
personal protective equipment (PPE)
pose recognition
Journal
IF:
8.9
Papers:
1.4W
Citations:
7.8W