Gait Analysis: Recognizing People by How They Walk

Gait analysis is a behavioral biometric that identifies people by the distinctive way they walk — stride length, pace, posture, and body sway. Because it can be observed from a distance without cooperation or physical contact, it is primarily used in surveillance and forensic video analysis rather than in point-of-access authentication.

Gait recognition systems typically analyze video footage frame by frame, tracking the movement of body joints and limbs over a walking cycle. Two broad technical approaches exist: model-based methods, which fit a skeletal or kinematic model to the walking figure and measure joint angles and stride timing, and appearance-based methods, which analyze the silhouette or overall shape changes of a person across a sequence of frames without explicitly modeling joints. The resulting features — cadence, step width, vertical body oscillation, arm swing — form a gait signature that can be compared against known profiles.

• Video capture from one or more camera angles, ideally a side profile
• Silhouette extraction or skeletal joint tracking across frames
• Feature extraction: stride length, cadence, posture, swing dynamics
• Comparison against stored gait signatures or clustering for anomaly detection

Gait analysis is most valuable in surveillance and forensic contexts: identifying a person of interest in low-resolution CCTV footage where the face is obscured or too distant to resolve, or corroborating identity in criminal investigations from security camera video. It also has applications in healthcare and rehabilitation for monitoring mobility disorders, and in some smart-building or smart-city research for tracking movement patterns for security or accessibility purposes.

The chief advantage of gait as a biometric is that it works at a distance and does not require the subject's cooperation, unlike fingerprint or iris scanning which need close proximity and often active participation. This makes it uniquely suited to covert identification scenarios. However, gait is far less stable than physiological traits: clothing, footwear, walking surface, fatigue, injury, and even mood can alter a person's gait, and accuracy tends to degrade compared to fingerprint or iris systems. It is generally used to narrow down candidates or add corroborating evidence rather than as a sole, definitive identifier.

Unlike iris recognition or fingerprint scanning covered elsewhere on this site, which require deliberate, close-range interaction with a sensor, gait analysis is passive and works from camera footage captured at a distance, much like early-stage facial recognition in surveillance settings. This makes gait a complementary tool for identification pipelines that combine multiple weaker signals — gait plus partial face plus clothing description — rather than a high-precision standalone authenticator.

Advances in computer vision and pose-estimation models have significantly improved the reliability of gait feature extraction from ordinary video, and researchers continue to explore combining gait with other passive signals for identification at range. Ethical and privacy considerations around covert, non-consensual identification remain a significant factor shaping how and where this technology is deployed.