Gait Recognition: Using Deep Learning to Collect Better Data

Gait recognition software is an emerging technology that garners a hodgepodge of reactions. To some, this technology is an overreach that’s used to surveil citizens and identify criminals in public. To others, this is a life-saving technology with massive implications for healthcare, particularly in the world of physical therapy. The latter understand that gait recognition software has improved the accuracy of gait analysis, which can prevent future injuries, speed up recovery times, and even unveil major underlying medical conditions.

In this article, we’ll explore how gait recognition software works, how it’s used for physical therapy patients, and why accurate data is so crucial to achieving positive outcomes.

What is gait recognition?

Gait recognition is the act of measuring the way someone walks or runs (known as their “gait”) in order to identify them. Thanks to advancements in machine learning, gait recognition has become incredibly sophisticated. One founder of this technology boasts a 94% accuracy rate in matching a person to their stride, and other sources claim accuracy as high as 98.4%. In that way, a person’s gait can be seen like their fingerprint and can be used in various ways.

One of the more uncomfortable applications of this technology is the way it has been used for surveillance. This tech is so state-of-the-art that it gives police the ability to measure a person’s gait through walls using Wi-Fi signals like radar. As a result of privacy concerns, some people are hesitant to support and encourage the development of gait recognition software.

But the truth is that this technology is far from being all doom and gloom.

The accuracy with which we can measure an individual’s gait is helping physical therapists around the world improve patient care, identify underlying conditions (such as Parkinson’s disease), and prevent falls/injuries among the elderly. And those benefits are due to the role that gait recognition technology (GRT) plays in improving the accuracy of gait analysis among physical therapy patients.

The difference between “gait analysis” and “gait recognition”

The terms “gait recognition” and “gait analysis” are similar but carry a big distinction in their overall purpose.

Gait recognition has a specific and narrow objective: to identify the user based on the unique characteristics of their gait. This is probably what you’ve read about gait recognition software in the news over the past few years. One of the most common examples is how Karnataka police identified the killer of journalist Gauri Lankesh by examining the suspect’s walk from CCTV footage. In the context of surveillance, gait recognition software is a neutral tool that’s been used for various societal and professional needs.

Gait analysis, on the other hand, is the evaluation of a person’s gait to look for proper form to improve recovery outcomes for patients, identify the risks for preventable falls, and spot serious conditions reflected in gait abnormalities. For example, measuring gait over time can be a reliable predictor of prodromal Parkinson’s disease, meaning the disease is present but hasn’t fully affected the patient’s motor functions.

So, where do the two concepts overlap?

The technology created by the former has greatly improved the accuracy of the latter. For example, authors of the article “Human Signature Identification Using IoT Technology and Gait Recognition” outline how interconnected the two ideas are in their abstract (emphasis added):

This study aimed to develop an autonomous design system for recognizing the subject by gait posture. Gait posture is a type of non-verbal communication characteristic of each person, and can be considered a signature used in identification. This system can be used for diagnosis. The system helps aging or disabled subjects to identify incorrect posture to recover the gait. Gait posture gives information for subject identification using leg movements and step distance as characteristic parameters.

The authors created a system for gait recognition and understood that the same system can be used to check for gait abnormalities in “aging or disabled subjects.” The improved accuracy of the gait analysis would help PTs spot larger underlying conditions earlier.

If gait analysis software shares some underlying mechanics with gait recognition software, should I have any privacy concerns?‍

The good news is that any reputable gait analysis app, like Exer Gait, will only use its measurements for analysis. These tools are not used for “fingerprinting” - in fact each analysis is done in a no-prior-knowledge environment to evaluate the current gait on its merits, rather than previous measurements. Those results can be compared to previous results by a PT or doctor, of course, but no identifiable gait profile is being developed with each evaluation.

Gait analysis apps don't care about who you are; instead, their exclusive aim is to evaluate gait at a point in time, and to give health professionals comparative data they can use to keep you healthy. Check out Exer's detailed privacy policy to better understand just how serious we are about protecting your data.

Let’s get a better idea of how this technology works.

How AI-driven technology measures gait

Machine learning software can measure gait more accurately than manual methods, but different models have been built depending on its application (single gait vs. multiple gait analyses, for example). Similar to other AI-driven software, though, the process relies on a powerful neural network and machine learning to track, model, and predict gait patterns. This often involves three components:

• A convolutional neural network (CNN)
• Machine learning
• Some form of sensory input (image, video, or data taken from physical sensors on the patient)

Without allowing ourselves to get too far into the weeds, let’s take a look at how these three components work together with a specific platform like Exer Gait.

👀 The Eyes

First, the patient can use any device with a 2D camera (like a smartphone, tablet, or laptop). We call this piece of the tech “The Eyes” because it’s this computer vision that observes the patient’s gait and movements in real time. This is the same technology you use to take pictures of your family vacation (i.e., the camera that comes standard with your smartphone or tablet).

This is an important point because it means PTs can run a gait analysis without the need for expensive, clunky, and non-transportable technology. In assisted living facilities, for example, imagine how difficult it is for patients with limited mobility just to get out of bed, let alone walk across the complex to a room with fixed tech. Now a PT can walk to the patient’s room with a mobile-friendly tool right in their pocket.

🧠 The Brain

When the patient walks in front of the camera, these movements get sent to the machine learning component of the app (or, as we call it, “The Brain”). This is what analyzes a patient’s form and “looks for” irregularities in gait rhythm, speed, and posture. It’s able to measure gait dysfunctions and provide real-time feedback to the user. But to classify a movement as either having “good” or “bad” form, the machine learning component needs an objective dataset on what proper gait looks like.

This is where our “Engine” comes into play.

⚙️ The Engine

Exer Gait is built on a convolutional neural network (CNN) that’s been trained on millions of custom datasets related to gait. Over time, the CNN learns what a healthy gait looks like so it can accurately understand human movement, poses, and exercises. So when a user measures their gait with any 2D camera, machine learning collects data on their movements and runs it through the CNN for analysis. Based on the millions of datasets already understood by the CNN, Exer can compare the patient’s real-time movements with what a healthy gait should look like. 

The app can then provide feedback to the user as it spots any gait irregularities.

Why is an accurate gait analysis important for maintaining good health?

Accurately measuring a patient’s gait is important for three reasons.

Improve patient outcomes

First, it can help improve outcomes for patients recovering from a specific injury. When a person walks, different body parts are used. Uneven distribution of weight can hinder the recovery process, especially if the original injury was a major lower extremity joint (ankles, knees, or hips). By improving gait form, the patient can avoid unnecessary strain on injuries that would otherwise slow down recovery.

Reduce the risk of future injury

Second, gait analysis can help PTs reduce the risk of future injury for the patient. This is particularly important for the elderly in assisted living facilities or who are at a higher risk of falls. An accurate gait analysis helps ensure patients are properly distributing their body weight and reducing movements that add unnecessary stress on muscles, joints, and bones.

Identify underlying health problems

Last, a gait analysis can help PTs identify warning signs of a serious underlying problem, such as cardiovascular disease, Parkinson’s disease, cancer, and even premature death. For example, a decrease in the range of a person’s swinging arms while walking can be an early indicator of Parkinson’s. This would be difficult for a PT to monitor without the help of tech because the human eye simply can’t compete with the accuracy of machine learning software.

Measuring gait securely and accurately

Gait recognition software is like every other piece of tech that’s been invented: it’s a neutral tool that can be used to accomplish different goals. The same technology that can be used to surveil peaceful protestors can also be used to catch dangerous criminals. In the context of healthcare, gait recognition software is very uncommon, while gait analysis is literally saving lives and extending good health outcomes for surgery patients and the elderly. Gait analysis plays a vital role in improving patient recovery times, reducing the risk for preventable injuries or falls, and identifying major underlying issues/diseases for early treatment.