Wi-Fi CSI-Based Outdoor Human Flow Prediction Using a Support Vector Machine.

Area of Science:

• Wireless Communications
• Machine Learning
• Human Activity Recognition

Background:

• Predicting human flow and activity is crucial for urban planning and traffic management.
• Current methods often rely on cameras, raising privacy concerns.
• Channel State Information (CSI) offers a privacy-preserving alternative for sensing human presence and movement.

Purpose of the Study:

• To propose a novel method for predicting human flow and activity in outdoor environments using CSI.
• To demonstrate the effectiveness of CSI in distinguishing various human activities like walking, running, and cycling.
• To achieve high prediction accuracy without compromising user privacy.

Main Methods:

• Extracting amplitude and phase components from measured CSI data.
• Calculating feature values including mean and variance of eigenvalues from auto-correlation and variance-covariance matrices.
• Utilizing a linear Support Vector Machine (SVM) classifier with leave-one-out cross-validation for prediction.

Main Results:

• The proposed CSI-based method achieved a maximum prediction accuracy of 100% for single-direction prediction.
• For two-direction prediction, the accuracy reached 99.5%.
• The method successfully differentiated between various activities including walking, running, and cycling for one, two, and three individuals.

Conclusions:

• CSI is a viable and effective technology for privacy-preserving human flow and activity prediction in outdoor settings.
• The proposed feature extraction and machine learning approach demonstrate high accuracy in recognizing diverse human activities.
• This research opens new avenues for intelligent transportation systems and ubiquitous sensing applications.