Gait Analysis | Filip Ciesinski

Gait Assessor

This project showcases my work that I have started in making a robust Gait analysis Model. It features a journal of my experience by diving into the research. Finally a visual showcase to demonstrate the application in detail.

Showing the Gait cycle in key parts from the (“Health & Gait: a Dataset for Gait-Based Analysis,” 2025)

Prior Research

Extracting pose maps is easy in this day and age. With all the tool sets available and the contemporary state of machine learning, I was able to prototype a novel edge based bottom-up pose estimator from Google's Mediapipe framework. If your interested check it out here on Github. This gave a real representation of what I have been looking for. Using this prototype I started learning the layers of depth that Gait Analysis requires. I am currently experimenting on a naive Temporal Segmentation Model attempting to predict the ground truth label on Gait characteristics. The datasets I'm using are multi-variant and controlled. As I get further on experimentation I plan to release all my findings and throw a hypothesis for pushing this area of research forward.

The current Research shows

This topic still needs more datasets in real world environments. However, the current point in progress of computer vision models shows an increase in reliability of computer vision models using Temporal and Spatial feature map representations. The current state of the art Open Gait Modeling approach has built several backbones to predict in more real world unconstrained environments. They have been able to do this by fusing 2 channels of perception. The first channel being a skeletal map also known as a pose estimator. This is the typical bottom up pose algorithm that represents key points within the skeletal frame of any human object. The current backbone for skeletal mapping is SkeletonGait. The second channel is shown to be a silhouette map which has been primarily used in prior models to extract full body shadows. The backbone for this model is DeepGaitV2. After each representation is extracted these are than fused to extract the average representation of both channels of perception into a multi-modal.

Data Set Breakdown

To demonstrate a visual of what data is used for this project. Check out my example below using data from (“Health & Gait: a Dataset for Gait-Based Analysis,” 2025).

Pose Skeleton

Acquiring gait metrics and anthropometric data is crucial for evaluating an individual’s physical status. Automating this assessment process alleviates the burden on healthcare professionals and accelerates patient monitoring. Current automation techniques depend on specific, expensive systems such as OptoGait or MuscleLAB, which necessitate training and physical space. A more accessible alternative could be artificial vision systems that are operable via mobile devices. This article introduces Health&Gait, the first dataset for video-based gait analysis, comprising 398 participants and 1, 564 videos. The dataset provides information such as the participant’s silhouette, semantic segmentation, optical flow, and human pose. Furthermore, each participant’s data includes their sex, anthropometric measurements like height and weight, and gait parameters such as step or stride length and gait speed. The technical evaluation demonstrates the utility of the information extracted from the videos and the gait parameters in tackling tasks like sex classification and regression of weight and age. Health&Gait facilitates the progression of artificial vision algorithms for automated gait analysis.

Gait Assessor

Prior Research

The current Research shows

Data Set Breakdown

Pose Skeleton

References

2025