Gait Recognition Systems Can Identify You with Your Manner of Walking

Gait recognition biometrics is a lesser known but a powerful biometric recognition method, in which subjects can be identified with their manner of walking. The theory behind this recognition system is that every person has a unique gait. It has also been a common experience that a familiar person can be recognized by his/her gait from a distance. Increasing influence of biometrics in today’s personal recognition needs has also led researchers to leverage capabilities of gait recognition. It is one of the few recognition methods that can identify people from a distance and can improve accuracy when used with other security and surveillance techniques.

Subsequent sections of this article discusses about gait recognition, how it works, gait recognition systems and their advantages / disadvantages.

• Physical factors: such as height, weight and physique of the person
• Intrinsic factors: person’s sex (M/F), age
• Extrinsic: clothing, terrain, footwear, etc.
• Pathological: diseases that can affect gait like musculoskeletal anomalies, neurological diseases, psychiatric disorders
• Physiological: proportions of body
• Psychological: emotions affecting the gait, personality type

Length of step, Length of stride, Speed, Cadence, Progression Line, Dynamic Base, Hip Angel, Foot Angle, Squat Performance are some of the parameters taken into account for while designing the system for the analysis of human gait.

Since gait is dependent on many factors, and each individual is being different, it is considered to be unique for each individual.

Gait recognition biometrics is the area of study in which human gait is taken as a biometric characteristics to uniquely identify individuals. Human beings have natural ability to recognize familiar people with their gait. Psychologists like Johansson G has shown that humans have ability to identify moving patterns in less than a second, which is not the case with static images showing a sequence of patterns.

This ability is developed over time by observing and learning to recognize patterns. On the other hand, a specific manner of walking is also developed as an individual age. It also keeps changing steadily with age. However, variations are slight when age is the only factor (except old age). It can also get affected by many other factors that we are going to discuss in a while. This coordinated, cyclic combination of movements that result in human locomotion is called gait. Each cycle of movements is repeated in the subsequent movements if physical or environmental conditions stay the same. These movements occur in a pattern and being dependent on many factors, they are considered to be unique for an individual. Not just manner of walking, but patterns of jogging, running and climbing stairs are also included in human gait.

A system designed to map human gait pattern and identify individuals on the basis of the mapped pattern is called the gait recognition system. Gait recognition biometrics is still in its infancy and different approaches are being worked upon. Every approach has its own set of advantages and disadvantages. A gait recognition system takes input from the data capture sub-system. It uses a capture device that can capture human motion. In a typical gait recognition setup, a video camera can be used for the purpose. It can also make use of on-body sensors, sensors on mobile phones or smart wearable devices, etc. to capture gait data. Radar based gait recognition systems are also being worked upon.

Complexity of a gait recognition system can depend heavily on the methodology leveraged. It can be a simple video camera based system in which captured video feed is analysed using the gait recognition software, while other techniques can make use of multiple sensors, infrared video camera, sensorized floor, etc. to capture more precise data of movement. Now when smartphones have become a commonplace and are carried around by people, some techniques have also been experimented, which make use of accelerometer data, a commonly found sensor on most smartphones.

As is the case with other biometric systems, efficiency of a human gait recognition system depends a lot on the recognition algorithm used. A gait recognition system may take inputs from multiple capture devices like video camera, sensors, wearables, etc. How efficiently the system can make use of this raw data depends on the efficiency of the recognition algorithm.

Data acquired by the gait data capture sub-system is further taken through different stages which are discussed later. Gait recognition algorithm used in the system processes this data (e.g. motion’s video feed or sensor data) to contour detection, silhouette segmentation and feature extraction. Algorithms used during features extraction stage can be crucial as it is where one gait pattern is distinguished from another. Raw data requirements of a gait recognition algorithm depend on the system type. For example, some algorithms may be designed to process only video feed, while others may require video feed as well as sensor data.

Machine learning is increasingly marking its presence in many computer based systems. Machine learning systems can improve themselves overtime by learning from data generated, captured and patterns. Gait recognition algorithms may have to deal with a lot of variations and their efficiency may be affected every time when there are unpredictable variations in raw data. Inclusion of machine learning in gait recognition algorithms can be the solution these system needs. They can learn and improve themselves every time they process a new sample of gait data.

Since no algorithmic approach can be absolute and there is always a scope for improvements, new capture methods and algorithms are always experimented upon to curb the inadequacies associated with the current gait recognition systems.

Amid increasing terrorist and criminal incidents, it has become very important that people entering or exiting a facility are accurately identified. It becomes more crucial when people are moving in a large group and frisking each individual is not possible. Public surveillance with biometrics like face recognition has been already deployed in many countries. China is on its way to create a country-wide network of surveillance cameras that are backed with face recognition and can locate a subject in a matter of minutes. Face recognition, however, has its own set of shortcomings. A suspect or criminal can cover his/her face to obstruct camera to capture facial details. An imposter can wear facial prosthetics or 3D masks to fool the system and bypass security.

These disadvantages can be addressed by deploying gait recognition along with other biometric recognition in for surveillance and security. Wherever other biometric recognition (e.g. face recognition) methods fall short, results of gait analysis can be compared to ensure that a subject is who the system says he/she is.

Many governments across the world run surveillance programs but do not officially disclose them due to privacy outcry. Running surveillance programs and deploying new security and surveillance technology has become an inevitable part of implementing national security. In the world increasingly getting insecure, biometric surveillance like gait recognition can do wonders. It does not require cooperation from the subjects and can surveil a large area quietly. The system can analyse manner of walking of each passer-by and compare with subjects on surveillance. Billions of dollars are spent on the name of security and surveillance every year, despite that, security incidents take place. Terrorist and criminal attacks result in loss of lives, disruption and people’s distrust from the government’s ability to control such incidents.

Biometric security and surveillance methods can do wonders if done right. Gait recognition may not have seen deployments as extensive as face or other recognition methods; however, it is increasingly getting attention of security experts to improve the rate of success in identifying subjects on surveillance.

The basic idea behind gait recognition is to equip a system with necessary hardware and software so that it can capture and map human gait to produce a digital signature that can later be compared against other human gait data to identify a subject.

The method chosen to capture gait data depends on the system’s requirements and objectives. Gait recognition systems designed for remote identification of subjects would require capture methods to be able to acquire data from a distance, while a system targeted for gait analysis of sportspersons, may use wearable sensors as well as video camera to capture gait data. Method of data capture may include one or more than one techniques that can capture the walking motion of a person.

Typically, a video camera is used to capture the motion when the subject needs to be identified remotely. In other applications, gait data can be captured using sensors in direct contact with the subject in motion, e.g. in sports biomechanics applications.

Other than camera and sensor based gait capture methods, researchers are also working on radar based gait recognition for remote identification of moving subjects. Instead of capturing video, the moving subject is showered with invisible radio waves. Since each individual’s walking style can be different, these radio waves bounce back little differently for each individual. The gait recognition system can figure out this difference and establish identity of a subject and use it to identify him/her in the future.

This stage takes place when gait data includes a video feed. Silhouette is a binary image extracted from the video feed of the moving subject. Silhouette segmentation makes an important part in implementation of machine vision. Extracting silhouette of a moving subject makes it easy to process and map as system does not have to deal with unnecessary 3D details. It also takes less processing power than processing 3D images.

In this stage, contours of the person are detected to specify the outer boundary of human body. There are several methods available for the purpose and choosing the right one depends on system objectives and other sub-systems like gait capture method used.

In these stages, gait features are extracted and finally, a classifier is used to identify a person. In the classification, the similarity between the extracted gait feature and the stored ones is computed to identify the walking person.

Just like any biometric system, gait recognition systems have their own set of advantages and disadvantages.

• Non-invasive biometric recognition makes is suitable for security and surveillance
• User consent is not required (and often not taken when deployed in mass / public surveillance applications)
• Monitoring people with gait recognition does not require their cooperation

• Its non-invasive nature is a threat to privacy but boon for security surveillance
• Still in its infancy, may not be adequate
• Many internal (e.g. diseases, psychological conditions, etc.) and external (e.g. clothing, footwear, etc.) may affect gait, hence the identification accuracy

At present, gait recognition system may not be accurate enough to handle a security or surveillance application alone, but they can help improve the accuracy when used with other biometric systems. They way security and surveillance is becoming an important aspect in everyday life, the day is not far when gait recognition systems will be watching your steps everywhere.

Human gait can be captured with video camera, sensors and even with radar waves and unique pattern can be extracted out of this gait data.

Human beings have some ability to identify people with their gait even if their face or other features are not visible. It becomes possible because people repeat a specific pattern while they walk. This pattern is developed over time and dependent on many physical and behavioral factors.  This manner of walking or gait can produce statistically significant data if mapped correctly and help recognize people.