Pedestrian Stride-Length Estimation Based on LSTM and Denoising Autoencoders.

Area of Science:

• Sensor Fusion and Signal Processing
• Human-Computer Interaction
• Robotics and Navigation

Background:

• Accurate stride-length estimation is crucial for applications like pedestrian dead reckoning and gait analysis.
• Existing algorithms struggle with accuracy in complex environments and natural human motion patterns.
• Inaccurate stride-length estimation leads to significant cumulative positioning errors in pedestrian navigation.

Purpose of the Study:

• To propose TapeLine, an adaptive stride-length estimation algorithm for smartphones.
• To automatically estimate stride length and walking distance using low-cost inertial sensors.
• To improve accuracy in complex environments and diverse motion patterns.

Main Methods:

• Developed TapeLine, integrating Long Short-Term Memory (LSTM) and Denoising Autoencoders (DAEs) for sensor data denoising.
• Utilized smartphone accelerometer and gyroscope data, along with extracted higher-level features.
• Created a data collection platform for simultaneous inertial sensor measurements, step events, stride length, and walking distance.

Main Results:

• TapeLine achieved a stride-length error rate of 4.63% and a walking-distance error rate of 1.43%.
• The algorithm demonstrated superior performance compared to state-of-the-art stride-length estimation (SLE) algorithms.
• Accurate estimation was achieved in diverse indoor/outdoor environments (stairs, escalators, elevators) and motion patterns (walking, running, jumping).

Conclusions:

• TapeLine offers a robust and accurate solution for stride-length and walking-distance estimation using smartphone inertial sensors.
• The algorithm effectively handles noise and complex motion, outperforming existing methods.
• It provides a practical, infrastructure-free solution for enhanced pedestrian navigation and activity recognition.