INIM: Inertial Images Construction with Applications to Activity Recognition.

Area of Science:

• Computer Science
• Machine Learning
• Signal Processing

Background:

• Human activity recognition (HAR) is crucial for applications like healthcare and indoor navigation.
• Smartphone inertial sensors are commonly used for HAR, with various machine learning approaches available.
• Existing methods include feature-based, 1D deep learning, and 2D deep learning architectures.

Purpose of the Study:

• To propose a novel framework for smartphone location and human activity recognition using inertial sensors.
• To leverage computer vision techniques for enhanced classification accuracy.
• To demonstrate the adaptability of the framework for diverse classification tasks.

Main Methods:

• A novel time series encoding approach converting inertial signals into 'inertial images'.
• Application of transfer learning from the computer vision domain to inertial sensor data.
• Utilizing two-dimensional deep learning architectures for classification.

Main Results:

• The proposed framework shows significant benefits across four diverse datasets.
• The inertial image encoding facilitates the use of established computer vision methods.
• Feature engineering is rendered redundant when using deep learning approaches.

Conclusions:

• The developed framework offers an effective method for smartphone-based activity and location recognition.
• The approach enables the application of powerful computer vision models to inertial sensor data.
• The framework is adaptable for other classification tasks involving inertial or similar sensory data.