A real-world indoor dynamic skeleton dataset of elderly abnormal gait including crutch-assisted walking
The paper introduces ElderlyPipe, a publicly released real-world indoor dataset of 63,855 quality-assured skeleton frames capturing elderly behaviors including crutch-assisted walking, which demonstrates that spatiotemporal graph models outperform single-frame approaches in analyzing gait imbalance.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Falls are a leading cause of injury and death for older adults, often resulting from subtle shifts in balance that go unnoticed until it is too late. To prevent these tragedies, caregivers and medical professionals need a way to monitor how people walk, looking for the early signs of instability before a fall occurs. For decades, researchers have tried to build computer systems that can watch a person and understand their movement, but these systems have struggled with one major problem: they lack good data. Most existing records of human movement come from young, healthy people in bright, controlled studios, or they rely on sensors worn on the body that people often forget to put on. What has been missing is a clear, detailed record of how elderly people actually move in the messy, unpredictable reality of a nursing home or a living room, especially when they are struggling to walk or using a cane for support. Without this specific kind of information, computers cannot learn to tell the difference between a normal step and a dangerous stumble.
A team of researchers from Zhejiang Sci-Tech University has now filled this gap by creating a new collection of data called ElderlyPipe. Instead of recording video of people's faces or bodies, which raises privacy concerns, the researchers captured movement and converted it into digital skeletons. These skeletons are not drawings, but rather a series of mathematical points that mark the location of a person's joints, such as shoulders, elbows, and knees, as they move through time. The team focused on recording real elderly residents in their daily environments, including hallways, lobbies, and dining areas, capturing four distinct behaviors: standing, walking, sitting, and the critical, unbalanced act of walking with a crutch. By gathering 63,855 of these high-quality movement frames from 287 continuous video segments, the researchers have provided a resource that allows computers to study the specific, shaky patterns of gait that signal danger in older adults.
The process of turning raw video into these useful digital skeletons required a careful, multi-step approach to handle the challenges of a real-world setting. In a nursing home, lighting can change suddenly, and people often block the camera's view of each other, causing standard computer vision tools to lose track of a person's limbs. To solve this, the researchers developed a system that first identifies the joints using a standard detection tool, then fills in the missing pieces when a limb is hidden. If a joint disappears for a short time, the system guesses its location by drawing a straight line between where it was seen before and where it reappears. If a limb is hidden for longer, the system uses the known length of the person's arm or leg to estimate where the missing joint must be, based on the position of the joints that are still visible. Once the skeleton is complete, the system smooths out the jittery, shaky movements that computers often introduce, creating a fluid and continuous record of the person's motion.
What makes this dataset particularly valuable is that it does not just show a person standing still; it captures the flow of movement over time. The researchers organized the data into small windows of twenty-four frames, allowing them to study how a person's body moves from one moment to the next. They found that looking at the sequence of movements together is far more effective than looking at single snapshots. When they tested computer models on this data, the system that understood the connection between past and present movements was much better at identifying the difficult, unbalanced gait of someone using a crutch. This system correctly identified these high-risk movements more often than models that only looked at one frame at a time, proving that the story of a fall is told in the rhythm of the steps, not just in the position of a single foot.
The data is designed to be safe and private, containing no images of faces or names, only the numerical coordinates of the joints. This allows the research community to build and test new safety tools without compromising the dignity of the elderly residents who contributed to the study. The release includes not just the data, but also the code and methods used to clean and organize it, ensuring that other scientists can reproduce the results. By focusing on the specific, often overlooked challenge of crutch-assisted walking in real indoor environments, this work provides a new foundation for developing technology that can one day watch over our elders, spotting the signs of imbalance and helping to keep them safe long before a fall ever happens.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.