Inertial Trajectory Estimation Using Low-Cost Inertial Measurement Units and Edge Computing.

Area of Science:

• Edge Computing
• Sensor Fusion
• Machine Learning for Robotics

Background:

• Trajectory estimation is crucial for applications like rehabilitation assessment and indoor navigation.
• Inertial Measurement Units (IMUs) offer a versatile solution for trajectory estimation across diverse environments.
• Existing methods often face challenges with transmission delays, privacy, and power consumption.

Purpose of the Study:

• To develop a secure, private, and low-power edge computing system for real-time trajectory estimation.
• To integrate advanced neural network models for enhanced motion tracking accuracy.
• To minimize transmission delays by processing data directly on an edge platform.

Main Methods:

• A novel trajectory estimation model was designed using Res2Net, a convolutional block attention module, and a temporal convolutional network.
• A motion dataset including walking and hand movements was collected for training and testing.
• The model was implemented on an edge computing platform with a neural processing unit.

Main Results:

• The proposed model achieved high accuracy with an average root mean-square error of 0.364 m.
• Inference time was significantly reduced to 0.234 s for 20s of IMU data, over 20% faster than a comparable model.
• The system demonstrated accurate real-time trajectory estimation capabilities on various edge platforms.

Conclusions:

• The developed edge computing system provides a secure, efficient, and accurate solution for real-time trajectory estimation using IMUs.
• The integration of advanced deep learning models on edge devices enhances performance for motion tracking applications.
• This approach overcomes limitations of traditional methods, enabling wider adoption in areas like personalized rehabilitation and autonomous navigation.