Reading time: 9 minutes
A person's gait is as unique as their voice timbre. Using this knowledge, gait recognition technology was created based on Machine Learning (ML) algorithms. ML-based systems can identify a person from an image even if their face is out of view, turned away from the camera, or concealed behind a mask.
The system analyzes the silhouette, height, speed, and walking characteristics and identifies the individual from a database. This technique is more convenient than retinal scanners or fingerprints in public places as it is unobtrusive. Moreover, gait recognition is unlikely to be outsmarted— every person's gait has no duplicates.
Table of Contents
- What is Gait Recognition? A Definition
- Gait Recognition Algorithm
- How Does Gait Recognition Work?
- Capturing Gait Data
- Silhouette Segmentation
- Contour Detection
- Feature Extraction and Classification
- What Do Gait Recognition Systems Analyze?
- How Gait Recognition Is Used
- How Accurate are Gait Recognition Systems?
- Advantage Of Gait Recognition
- Disadvantages Of Gait Recognition
- How do gait biometrics work?
- Is gait recognition accurate?
- What features are observed in gait recognition?
- How can gait recognition be fooled?
What is Gait Recognition? A Definition
Gait is an important indicator that is used in behavioral biometrics to identify a person over a long distance without direct contact. When a person walks, it is possible to observe around 24 individual parameters and movements (https://dl.acm.org/doi/pdf/10.1145/3230633) that form the uniqueness of the gait.
With the development of Computer Vision (CV) techniques, there are many approaches to human identification by movement in video, using natural biometric characteristics (the human skeleton, silhouette, change while walking) and abstract features.
A gait recognition system uses the shape of the human body and the way it moves in order to identify it. The software, using CV algorithms, detects a human silhouette on video and analyzes its movements. These data create a human behavioral model.
Gait Recognition Algorithm
Gait recognition technology uses several sources or capture devices— video cameras, motion sensors, and so on— to acquire data. The acquired data go through a number of recognition steps.
The main algorithm recognizes the gait, processes the received data, detects contours, silhouettes, and segments individual human features. Then, the feature extraction algorithm comes into effect— this distinguishes one gait from another.
These algorithms can vary, and their requirements can also be different. For example, some algorithms are designed to process video signals, while others use data from sensors.
Thanks to machine learning technologies, scientists can improve recognition systems based on the data and models obtained. Because each gait is unique, the recognition algorithms encounter new data every time they are used. The more gait variants the system sees, the better it analyzes future data.
So, suppose the algorithm processes two very similar gaits. The pattern recognition and silhouette segmentation algorithms are trained to separate the subtle details and enter them into the database. This allows for better gait classification and better results going forward.
How Does Gait Recognition Work?
Researchers compiled a database of individuals' gaits, including around 20,000-foot movement recordings from 127 people, taken with special floor sensors and high-resolution cameras.
All these data were loaded into a neural network for image processing. After training, it recognized people by their gait with close to 100% accuracy. The work of the ML system is based on the principle of deep residual learning, enabling the identification of a person through the spatial and temporal characteristics of their footprint.
The most common gait recognition system is based on four components:
- capturing gait data;
- silhouette segmentation;
- contour detection;
- feature extraction and classification.
Capturing Gait Data
Video cameras or wearable sensors can be used to capture a gait. The most striking example of such sensors may well be the special costumes that actors wear on set so that motion artists can later draw the character based on their movements.
Another method of gait capture involves the use of radar to identify moving objects remotely. The object of interest is irradiated with radio waves that reflect off their body. The system recognizes the reflected waves and uses the data for identification.
This stage is appropriate for studies that use video camera recordings. A binary image of a person's silhouette is extracted from the recording and studied by vision-based algorithms. Silhouette segmentation makes it easier for the algorithm to process and map a complete picture.
Next, the system defines the boundaries of the human body— highlighting the contours. The methods used to achieve this goal may vary depending on what hardware (cameras or wearable sensors) is used to capture the gait.
Feature Extraction and Classification
In the final step, the individual features of a gait are determined. Here a classifier is used to identify the person, which is then entered into a database and used for detection.
What Do Gait Recognition Systems Analyze?
Gait biometric systems capture step patterns using video images and then convert the collated data into a mathematical equation. Gait as a biometric measure can be influenced by several factors, including footwear, terrain, fatigue, and injury.
The primary determinant of gait is the size of the human skeleton. Terrain also plays a role— it can cause changes in a person's speed. Injuries and footwear are also significant— if a person walks barefoot, their gait changes; if a person has been injured, the effect on the gait can be unpredictable. In addition, the system needs to analyze the development of a person's muscles and how tired they are.
How Gait Recognition Is Used
Gait recognition systems are mainly used in video surveillance and come in handy in crowded areas for security purposes. The systems can quickly spot a person who is wanted by the authorities and prevent terrorist attacks or other forms of crime.
A new gait recognition technology was introduced by the Chinese company Watrix. The developers claim that the system identifies people in videos by silhouette and gait from a distance of 50 meters. The accuracy of this technology is up to 94%.
But gait recognition in fields other than just security. For example, this technology can become an element of the smart home ecosystem. It could also be used in nursing homes to alert staff if a patient falls. In hospitals, the system can help diagnose neurological disorders and plan rehabilitation therapy, and athletes will find it helpful in training.
How Accurate are Gait Recognition Systems?
Scientists at the University of Manchester in the United Kingdom have developed a high-precision human gait recognition system. Researchers believe it could replace retinal scanners or fingerprints in the future. When a person walks, the system tracks around 24 different parameters and movements that form the uniqueness of the gait, notes study leader Omar Costilla Reyes.
To create a system known as SfootBD, scientists collected a database of twenty thousand signals received from 120 people while they were walking. Each person's gait was measured using special pressure sensing floor panels and a high-resolution camera. Scientists then analyzed weight distribution, walking speed, and some three-dimensional measures of each gait.
The researchers tested the development in three situations: at airport checkpoints, in the workplace, and at home. The system can recognize a person it already knows with almost 100 percent accuracy, and the error rate was just 0.7 percent. The system also copes successfully with cheating— when someone tries to imitate a gait.
The SfootBD system is almost 380 times more accurate than previous developments by other researchers: in one of them, test subjects had to walk barefoot on special touch panels, and in the other, scientists used 3D-shooting technology, which compared the way people walk with surveillance camera footage.
Advantage Of Gait Recognition
Gait recognition technology is less “touchy-feely” than other biometric verification systems such as retinal scans or fingerprints. Thus, it is non-invasive and can be applied without user consent. Moreover, the success rate of this technology is high— the error rate is only 0.7%.
Disadvantages Of Gait Recognition
The new technology has its limitations: it requires special sensing panels and a high-resolution camera. Also, the system can only recognize people whose data have been recorded in advance and stored in its database, so it does not yet have a wide range of applications.
The possibility that it could be used covertly without people's knowledge is also troubling, which raises concerns about security and privacy.
In addition, since the technology is still in its developmental stage, its results cannot always be trusted 100%. For example, many internal factors (e.g., disease, psychological conditions, etc.) can affect gait much more than external factors and impair recognition accuracy.
Of course, the new system is not universal. First, to reconstruct a person's gait and enter it into the database, you have to put them in a room with sensors on the floor and monitor them with high-resolution cameras. Secondly, the algorithms recognize only those in the database, and therefore scaling the technology will be rather problematic.
On the other hand, the technology requires no subject contact and is easily customizable for use in public places. In addition, the error rate of the new algorithm is negligible, which means that, with proper development, such a system could be very promising.
Now let us take a look at some of the most popular questions regarding gait recognition technology.
How do gait biometrics work?
Neural networks can find patterns in a person's gait, which can then be used to recognize and identify individuals with almost one hundred percent accuracy. The gait is digitized using high-resolution sensors and cameras. The system then analyzes all the data, including weight distribution, gait speed, and three-dimensional features of each walking style. The result is a fully-fledged «cast» of the person and their walk, which is entered into a database for use in the future.
Is gait recognition accurate?
The results of the experiments showed that this technology recognizes a person with almost 100% accuracy— the error rate was just 0.7%. Thus, we can say that the percentage of accuracy of gait recognition technology is quite high. Nevertheless, since the technology is still under development, it is too early to draw any strict conclusions.
What features are observed in gait recognition?
Gait recognition systems capture step patterns using video images. Several factors are required for accurate recognition. Among them are footwear, terrain, fatigue, injuries, and the person's muscle development.
How can gait recognition be fooled?
As long as the technology is still in its infancy, it is probably still possible to cheat— but you have to try hard. The system copes successfully when someone tries to imitate the gait of another person. After all, each person's gait is unique, and they're not as easy to fake as it might seem.