Visual–Inertial Fusion Framework for Isolating Seated Human-Body Vibration in Dynamic Vehicular Environments

TL;DR

This paper introduces a new system that uses visual and inertial data to study how vibrations from vehicle seats affect the human body and how people adjust their posture.

Contribution

A novel visual–inertial fusion framework is introduced to isolate and analyze seated human-body vibration in vehicles.

Findings

The framework can distinguish passive ride phases from strongly compensated phases.

It separates camera jitter from true body motion and reveals anisotropic postural strategies.

Abstract

Understanding how seat-induced whole-body vibration (WBV) is transmitted to and actively compensated by the human body is essential for accurately assessing discomfort, fatigue, and postural control in vehicle occupants. This study proposes a visual–inertial fusion framework utilizing IMU-RGB-D data to isolate seated human body vibration in dynamic vehicular environments. In real-cabin monitoring systems, measured motion is a superposition of platform vibration, passive transmission through the body, active postural compensation, and camera jitter. Existing WBV and driver monitoring studies typically rely on single modality sensing, such as inertial or visual approaches, without decomposing these components or modelling camera vibration. The framework synchronized three IMUs with RGB-D landmarks. Seat, human body, and camera accelerations are separated, and body vibration velocity is derived from body–seat differential acceleration via band-pass filtering and spectral integration. The 3D landmarks enable rotational-translational Postural Compensation Index metrics, axis-wise energy distributions, and anthropometric consistency checks. The study is held in an in-service urban tram case. Torso vibration is dominated by 40% anteroposterior components, while head postural is predominantly > 50% lateral sway. Near static anthropometric evaluation was also studied, resulting in shoulder width errors that remain within ±10–20 mm. The results show that the framework can distinguish passive ride phases from strongly compensated phases, separate camera jitter from true body motion, and reveal anisotropic postural strategies, providing a structured basis for vibration and posture analysis in in-vehicle monitoring.

Figures