Identification of a Person in a Trajectory Based on Wearable Sensor Data Analysis

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Wearable Technology

Background:

• Tracking human movement using edge device cameras is essential for various applications.
• Matching camera-derived trajectories with wearable sensor data (acceleration, angular velocity) presents a modality challenge.
• Incomplete trajectory data complicates individual identification.

Purpose of the Study:

• To develop a robust method for synchronizing and matching human trajectories from cameras with wearable sensor data.
• To address the challenge of fragmented and incomplete trajectory information.
• To improve the accuracy of identifying individuals based on multimodal data.

Main Methods:

• Proposed the SyncScore model to quantify similarity between trajectory segments and sensor data.
• Developed a Likelihood Fusion algorithm for systematic, time-integrated similarity updates.
• Utilized the UEA dataset for model validation.

Main Results:

• The proposed models achieved an accuracy of 0.725.
• Sensitivity and F1-score were also confirmed to be 0.725.
• Demonstrated effective matching of human trajectories and sensor data.

Conclusions:

• The SyncScore model and Likelihood Fusion algorithm offer a viable solution for multimodal human movement analysis.
• The method shows promise for applications requiring accurate individual identification from partial data.
• The approach achieved competitive results on a standard benchmark dataset.