Recognition of Walking Terrain Using Machine Learning

This paper presents a machine learning-based approach for classifying walking terrains using accelerometer and altimeter data. The system was implemented using three different models: an Artificial Neural Network (ANN), a Long Short-Term Memory (LSTM) network, and a 1-Dimensional Convolutional Neural Network (1D-CNN). Data was collected from an Arduino Nano 33 BLE device, and various preprocessing techniques, including feature extraction, were applied to enhance performance. The LSTM model achieved the highest accuracy at 96.00%, outperforming both the 1D-CNN model (95.67%) and the ANN model (83.98%). The LSTM's superior performance is attributed to its ability to capture temporal dependencies in time-series data, particularly improving classification on dynamic terrains such as stairs. The 1D-CNN also performed well, leveraging convolutional filters to identify local patterns in the data. This work demonstrates the potential for deploying low-cost, real-time terrain classification systems in wearable devices, with applications in fitness tracking and healthcare monitoring. Future research will focus on expanding activity recognition capabilities and integrating the models into wearable technologies.