Gait Recognition

1
Gait Recognition

• Simon Smith Jamie Hutton
• Thomas Moore David Newman

2
Outline

• Gait?
• Enrolment
• Symmetry
• Hu Invariant Moments
• Classification
• Conclusions
• Demonstration

3
Gait?

• From Old Norse gata for path
• But possibly from Northern derivative of goat
• Why use gait as a biometric?
• Non-invasive
• Process sequence of images
• More information than other biometrics
• Greater robustness/reliability
• Gait recognition methods
• Model-based
• Holistic approach

4
Holistic Methods

• Chose to implement two holistic approaches
• Less computationally complex, faster
• More suitable for online demonstration
• A simple representation of Gait
• Raw numbers, images
• Problems with occlusion, noise

5
Enrolment

• Capture subjects gait
• Video ideally with chroma-key background
• Avoid occlusion of subject
• Outdoor images cause some problems
• Process video of subject walking
• Background subtraction
• Indoor Chroma-key, Outdoor Mixture of
  Gaussians
• Binary silhouette of each frame
• 30 frames captures complete gait cycle
• Begin at known heel-strike

6
Symmetry

• Crop and resize images to 64x64
• Centre the body in the image
• Extract symmetry for each image in sequence
• Average all symmetry maps to get Gait Signature
• Compare Gait signatures directly

Number of images
7
Hu Invariant Moments

• Shape descriptor, combines moments to give
  invariance to
• Rotation
• Translation
• Scaling
• Originally designed for single shape description,
  extended here for sequences
• We use Hu1, Hu2 and Hu8 moments
• Other moments fail to discriminate between
  subjects

8
Classification

• k Nearest Neighbours classification
• Euclidean distance
• Up to 6-dimensional feature space
• Mean of 1 or all Hu moments
• Variance of 1 or all Hu moments
• Tie Resolution
• Highest ranking matches chosen

9
Conclusions

• Evaluation of two Holistic Gait descriptors
• Hu Moments
• Good indoor performance
• Poor performance outdoor
• Needs higher dimensional parameter space
• Ability to ignore/correct anomalous results
• Symmetry
• Good indoor, better outdoor performance
• Larger population may cause poor performance

10
Demonstration