Transportation Mode Detection Using Temporal Convolutional Networks Based on Sensors Integrated into Smartphones

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Accurate transportation mode detection is crucial for intelligent mobile services and human activity identification.
• Existing methods face challenges in achieving reliable and precise transportation mode detection.
• Advancements in sensor technology and deep learning offer new possibilities for improved detection.

Purpose of the Study:

• To propose a novel algorithm, T2Trans, for accurate transportation mode detection.
• To leverage temporal convolutional networks (TCNs) and multi-sensor data for enhanced feature learning.
• To improve the efficiency and accuracy of identifying various transportation modes.

Main Methods:

• Developed T2Trans, a transportation mode detection algorithm based on temporal convolutional networks (TCNs).
• Utilized data from multiple lightweight sensors integrated into a smartphone.
• Employed feature representation learning on preprocessed sensor data using TCNs.

Main Results:

• Achieved a macro F1-score of 86.42% on the SHL dataset and 88.37% on the HTC dataset.
• Obtained an average accuracy of 86.37% on the SHL dataset and 89.13% on the HTC dataset.
• Demonstrated superior performance compared to benchmark algorithms in identifying eight transportation modes.

Conclusions:

• The T2Trans algorithm effectively improves transportation mode detection accuracy and learning efficiency.
• TCNs applied to multi-sensor data provide robust feature representation for transportation mode identification.
• The proposed method offers a reliable solution for intelligent mobile services requiring accurate human activity recognition.