1 / 24

A Review Of Gait Recognition Results In Psychology (The Origin Of The Vision)

A Review Of Gait Recognition Results In Psychology (The Origin Of The Vision). Professor Kevin W. Bowyer Computer Science & Engineering University of Notre Dame. Two Important Papers .

montana
Download Presentation

A Review Of Gait Recognition Results In Psychology (The Origin Of The Vision)

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. A Review Of Gait Recognition Results In Psychology (The Origin Of The Vision) Professor Kevin W. Bowyer Computer Science & Engineering University of Notre Dame Copyright, Kevin W. Bowyer, 2001

  2. Two Important Papers • Gunar Johansson, “Visual Perception of Biological Motion and a Model For Its Analysis,” Perception and Psychophysics 14, 201-211, 1973. Introduced and popularized recognition of human gait from video of points fixed at body joints (“point-light displays”).

  3. Two Important Papers 2. L.T. Kozlowski and J.E. Cutting, “Recognizing Friends By Their Walk: Gait Perception Without Familiarity Cues,” Bulletin of the Psychonomic Society 9, 353-356, 1977. Advanced and popularized the idea that you could recognize the specific person from the appearance of their gait, using PLDs.

  4. One Other Paper L.T. Kozlowski and J.E. Cutting, “Recognizing the Sex of a Walker From a Dynamic Point-Light Display,” Perception and Psychophysics 21, 575-580, 1977. Advanced and popularized the idea that you could classify the sex of the walker from the appearance of their gait, using PLDs.

  5. Johansson’s Work The “point-light display” – • A set of bright points moving against a dark background. • Points are shoulders, elbows, wrists, waist, knee, and ankles of the subject. • Person is viewed as the set of points while walking some straight path.

  6. Johansson’s Work Demonstration 1 – • Observers view PLD on video monitor. • Subject walks fronto-parallel to camera. • “always evokes the same spontaneous response after the first one or two steps.” • “Perceiving a walking person in the motions of these 10 spots seems to be equally as … natural as seeing a real …”

  7. Johansson’s Work Demonstration 2 – • Like Demonstration 1, but subject walks toward camera at angle. • “All observers immediately reported seeing a person walking toward them.” • “perceived direction of motion track roughly corresponds to the recorded one.”

  8. Johansson’s Work Demonstration 3 – • Like Demonstrations 1 and 2, but with only five points seen: waist, knees, ankles. • “Observers always report seeing a walking human being.” • “In most cases, … as two walking legs.” • “A few … the upper point as shoulders.”

  9. Johansson’s Work “a number of other demonstration patterns have also been produced and studied. … running …, cycling, climbing, dancing in couples, various types of gymnastic motion, etc. In all of these cases, spontaneous and correct identification of the types of activity has been made without exception.”

  10. Johansson’s Work Experiment 1 – • Display viewed without horizontal motion component; subject is walking “in place” for about 10 steps. • 10 subjects, not seen PLDs before, “neutral instruction to describe what they saw.” • “All subjects immediately reported seeing a walking person.”

  11. Johansson’s Work Experiment 2 – • Like Experiment 1, but viewing only 1.0 second of display, or about one step. • Again, 10 naïve subjects. • “After the first exposure, all subjects, without any hesitation, reported seeing a walking human being.”

  12. Johansson’s Work Experiment 3 – • Like 1 and 2, but with extra circular motion added by a rotating mirror; 3 to 4 steps. • Again, 10 naïve subjects. • “All subjects immediately reported seeing a walking man. … said that the walker moved in a highly strange, ‘wavy’ way.”

  13. Johansson’s Work “The demonstrations described have in a conclusive way made clear that 5-10 elements in adequate combinations of proximal motion give the visual system highly efficient information about human motion.”

  14. Johansson’s Work “We can, with a high degree of confidence, conclude that it was not previous learning of motion patterns which determined the perception of walking in our Experiments 1-3. Instead, we have found that it seems to be a highly mechanical, automatic type of visual data treatment that is most important.”

  15. Johansson’s Work Very well known work; the PLD was seen as impressive, results were striking, demo video was apparently widely circulated. Many people may remember this work as dealing with identification of specific people, but it is really concerned with perceiving that a generic person is depicted in the stimulus.

  16. Kozlowski & Cutting Recognizing friends by their walk: • Six subjects (3 M, 3 F) with “normal” gait; knew each other, in university housing. • Each subject walks 10 times, about 5 strides in each time. • Same six served as viewers a month later, plus one other person that knew the six.

  17. Kozlowski & Cutting “Our viewers did well, although they were far from perfect. Chance performance would be 16.7% correct identification. Overall, correct responses occurred on 38% of trials. The range of performance was 20% to 58%…”

  18. Kozlowski & Cutting “Although no feedback was given, performance improved over the course of the task. Viewers increased from 27% correct identification for the first three presentations of each of the six walkers to 59% for the last three.”

  19. Kozlowski & Cutting “The confidence judgments reflect the fact that viewers knew what they were doing. When maximally confident, they were 75% correct.” “Self-recognition was not better than recognition of others (43% versus 36%).”

  20. Kozlowski & Cutting “Viewers tended to mention as clues certain critical features of the display, such as the speed, bounciness, rhythm of the walker, amount of arm swing, or the length of steps. The most accurate viewers claimed to associate these dynamic aspects of the display with particular individuals.”

  21. Kozlowski & Cutting “It is our impression that performance would continue to improve without feedback, and that one or two trials with feedback for each walker would yield essentially perfect performance.”

  22. Kozlowski & Cutting Note that there is no variation in: • Walking surface • Shoe type • Clothing • Time

  23. Kozlowski & Cutting #2 • Recognizing the sex of a walker: • Very similar to other paper • “Walkers 4, 5, and 6 were correctly identified as male on an average of 72% of trials, and walkers 2 and 3 were correctly identified as female on 67% of trials … Walker 1 … only 32% of trials.”

  24. Summary • Evidence that humans can recognize their friends from a sparse representation of gait does exist, but is “thin” when considered in the context of the Human ID program. • Six subjects under highly controlled conditions achieved only about 75% correct identification. More people and/or more variation would drive the result down.

More Related