This post is also available in:
עברית (Hebrew)
Identifying people from a distance has long been a challenge for security systems. Facial recognition often requires high-resolution images and unobstructed views, while clothing, camera angle, or partial occlusion can reduce the effectiveness of many existing identification methods. As a result, researchers are exploring alternative biometric characteristics that remain recognizable even when facial features are unavailable.
A newly developed artificial intelligence system takes a different approach by identifying individuals based on the unique way they walk.
Instead of analyzing body appearance, the system focuses on gait, as in the distinctive movement patterns created as a person walks. Every individual moves their joints slightly differently, creating a biomechanical signature that can be measured and compared over time.
The AI platform extracts key points representing major body joints from ordinary video footage. It then calculates the position of those joints as well as their angles, angular velocity, and acceleration throughout each walking cycle. Together, these measurements form a detailed digital representation of a person’s gait.
According to TechXplore, to improve accuracy, the researchers separated body posture and movement into two independent processing streams before combining the results. This allows the system to analyze how a person is built and how they move as complementary sources of information rather than treating them as a single dataset.
Another notable feature is an attention mechanism that dynamically prioritizes different body parts depending on what the camera can see. For example, if a person’s legs are partially hidden behind an obstacle, the AI automatically shifts greater emphasis toward arm movement and other visible joints instead of discarding the observation.
The researchers report that the system maintained reliable performance despite variations in clothing, camera perspective, and partial obstruction, which are conditions that often reduce the effectiveness of conventional gait recognition techniques.
From a defense and homeland security perspective, long-range gait recognition could complement existing surveillance and identification systems. Border security, critical infrastructure protection, military base security, and law enforcement operations often require identifying individuals before facial imagery becomes available. Because the technique relies primarily on movement rather than facial appearance, it may also reduce the amount of personally identifiable visual information that must be processed.
Testing on several public gait-recognition datasets showed the new approach outperforming existing methods.
As AI-driven biometrics continue to evolve, gait recognition is emerging as another tool that could strengthen long-range identification while expanding the range of conditions under which automated security systems can reliably recognize individuals.
The research was published here.
