Understanding the Representation of Older Adults in Motion Capture Locomotion Datasets

1 st Yunkai Yu, 2 nd Yingying Wang, 3 rd Rong Zheng · 2025

Read Understanding the Representation of Older Adults in Motion Capture Locomotion Datasets by 1 st Yunkai Yu, 2 nd Yingying Wang, 3 rd Rong Zheng (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-11713.txt.

Introduction

Yunkai Yu, Yingying Wang, and Rong Zheng audit how well older adults are represented in motion capture datasets of human locomotion. These datasets supply ground truth for activity recognition, pose estimation, and fall detection, all technologies aimed heavily at older users.

Forty-one public datasets were surveyed. Only eight include motions actually performed by older adults, and four more use younger actors performing movements labeled as old style. That thin base is worrying, because models trained on young bodies may misread the movements they most need to catch, including falls. The survey gives dataset builders a clear account of a representation gap with safety consequences.

Catalog record

CREATOR1 st Yunkai Yu, 2 nd Yingying Wang, 3 rd Rong Zheng
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-11713

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Understanding the Representation of Older Adults in Motion Capture Locomotion Datasets Thanks: This work is supported partly by NSERC sMAP CREATE and NSERC Discovery grants. 1 st Yunkai Yu, 2 nd Yingying Wang, 3 rd Rong Zheng Abstract The Internet of Things (IoT) sensors have been widely employed to capture human locomotions to enable applications such as activity recognition, human pose estimation, and fall detection. Motion capture (MoCap) systems are frequently used to generate ground truth annotations for human poses when training models with data from wearable or ambient sensors, and have been shown to be effective to synthesize data in these modalities. However, the representation of older adults, an increasingly important demographic in healthcare, in existing MoCap locomotion datasets has not been thoroughly examined. This work surveyed 41 publicly available datasets, identifying eight that include older adult motions and four that contain motions performed by younger actors annotated as old style. Older adults represent a small portion of participants overall, and few datasets provide full-body motion data for this group. To assess the fidelity of old-style walking motions, quantitative metrics are introduced, defining high fidelity as the ability to capture age-related differences relative to normative walking. Using gait parameters that are age-sensitive, robust to noise, and resilient to data scarcity, we found that old-style walking motions often exhibit overly controlled patterns and fail to faithfully characterize aging. These findings highlight the need for improved representation of older adults in motion datasets and establish a method to quantitatively evaluate the quality of old-style walking motions. I Introduction The Internet of Things (IoT) has made locomotion data more accessible, facilitating innovations in applications ranging from human activity recognition to fall detection and exoskeleton design. A motion capture (MoCap) system is the gold standard for recording locomotion data with high precision in controlled lab environments. MoCap data has been widely used to annotate data from IoT sensors such as mmWave radars and inertial measurement units in human motion analysis [ 58 ] , or synthesize data from other modalities [ 18 ] . Data-driven models are increasingly important in motion analysis, often trained directly on MoCap datasets without relying on hand-crafted features. As such, the characteristics of these datasets, such as the variety of motions and the demographic composition, directly influence model performance. Considering the growing older population worldwide, it is important to understand the coverage and quality of old adult motions in public datasets. David et al. [ 8 ] surveyed 93 locomotion datasets and reported number of subjects, motor skills, application domains, sensory systems, and data formats. They found that existing datasets lack detailed demographic information of the participants. Few studies have systematically examined how well existing MoCap datasets capture the motion patterns of older adults. This work aims to answer two research questions. 1) What is the availability of older adult locomotion data in existing public MoCap datasets? and 2) Do existing old-style locomotion data performed by actors have sufficient fidelity to accurately reflect age-related differences? To answer the first question, we surveyed 41 datasets collected by clinical and engineering communities and analyzed their demographic coverage, motor skills, motion varieties, and whether the data includes full-body or partial-body motions. Among these, only eight datasets are known to include older adults. In total, 121 of the 385 participants are older adults, but full-body motion data is available for 75 of them. The motor skills in full-body motions are limited to walking and standing for older adults. To answer the second question, we further analyzed the old-style forward walking motions in four of the 41 datasets. We selected gait parameters known to capture age-related differences and used them as quantifiable metrics to evaluate the fidelity of old-style motions. Our analysis reveals that only around 12 minutes of old-style forward walking are available, and the recorded movements often fail to accurately represent age-related gait patterns. In summary, the main contributions of this work include: • It conducted the first survey of public MoCap locomotion datasets focusing on demographic coverage, motor skills, motion varieties, and the anatomical completeness of MoCap for older adults. • It proposes key objective metrics for assessing the fidelity of old-style motions with respect to their ability to reflect age-related differences. • It systematically reveals for the first time the low availability of high-quality locomotion data for older adults in public datasets. The rest of the paper is organized as follows. Section II introduces key terminologies used in subsequent discussions. Section III presents a qualitative analysis of the datasets discussed in this paper. Section IV presents our quantitative assessment method and assesses the fidelity of old-style motions in existing MoCap locomotion datasets. Section V concludes the paper with key findings and future directions. II Preliminaries Clinical datasets in this work are interested in locomotion profiles for specific groups. In these datasets, motion data is collected from participants of interest in controlled lab environments. In addition to MoCap data, some datasets, e.g. [ 49 ] , include ground reaction force and electromyography (EMG) data to understand human dynamics. General-purpose datasets are mainly collected to support applications in video games, animation, simulation, and other fields that rely on human motions. These datasets aim to capture a wide range of motor skills, enabling researchers and practitioners to create realistic and adaptable animations. Typically, the motions in these datasets came from campus volunteers or professional actors, who are directed to perform various motor skills with precision and consistency. Motor skills refer to the ability to perform specific functional movements that involve coordination of muscles. Style is the manner in which a motor skill is performed. It conveys emotions, personality, and intentions. For example, Xia et al. ’s dataset [ 54 ] includes eight distinct styles: angry, childlike, depressed, neutral, old, proud, sexy, and strutting. In contrast, Ian et al. [ 31 ] introduced the 100style dataset, which contains 100 performative styles, with old being one of them. The BFA [ 2 ] dataset also guided actors to perform motions to resemble older adults. Motions labeled with such style annotations are referred to as stylized motions. Old style is subjective and can be performed using different movement patterns depending on the performers’ interpretation. Motion variety refers to the range and diversity of movement patterns. In this work, we focus on natural and age-related movement patterns within the same motor skills, excluding artificial variations sourced from unnaturalness or age-irrelevant factors. Ambiguity in the guideline and procedure designed to collect motions results in variations even within the same motor skill, which are excluded from consideration. For example, the motion description of the 0005{\_}{Walking001} in the SFU MoCap dataset [ 24 ] is “walking forwards”. However, the walking motion clip includes instances of turning around. Similarly, a clip labeled as walking in straight lines with increasing stride length includes an intentional modification of gait dynamics, which constitutes an artificial variation rather than a natural difference [ 24 ] . Moreover, participants may not have rigorously followed the same experimental setup when performing motions in different styles. For example, text annotations indicate that sequences 137\_{33} and 137\_{29} in the CMU MoCap dataset [ 7 ] correspond to the same motor skill. However, their walking trajectories differ significantly. III

Read the complete text