html5-img
1 / 26

A Novel Sequence Representation for Unsupervised Analysis of Human Activities

A Novel Sequence Representation for Unsupervised Analysis of Human Activities. Presented by: Wei Pan For CS88/188. The Unsupervised Activity Classification System. Length . A 40-page paper. Straight-forward way of thinking of a problem. No graph model, no inference, no fancy math.

tamah
Download Presentation

A Novel Sequence Representation for Unsupervised Analysis of Human Activities

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 Novel Sequence Representation for Unsupervised Analysis of Human Activities Presented by: Wei Pan For CS88/188

  2. The Unsupervised Activity Classification System

  3. Length • A 40-page paper. • Straight-forward way of thinking of a problem. • No graph model, no inference, no fancy math.

  4. Definition • Key Object • Fridge, washer, stove, sink… • Event • Interaction among a subset of key objects in a certain time. (turn stove on; eat egg; fry egg) • Activity • A sequence of events with temporal order. • Activity Structure • The event sequence of an activity.

  5. Definition • n-gram Histogram • An Activity could be represented by a subset of its sequence.

  6. Definition • Is n-gram real work? • Under certain assumption it works! • Simulation with VMMC. • VMMC: A sampling method (in this paper) to generate sequences of different classes with noise.

  7. Unsupervised Classification • Distance Measurement. • Clustering Algorithm. • Cluster Modeling.

  8. Problem #1 • Distance between two activities? • Y, Z are events in A and B respectively. K is normalization factor.

  9. Problem #2 • Clustering Algorithm • A max clique is a class. • Dominant set algorithm. ([Pavan2003])

  10. Problem #3 • Each activity is one of the two types: • Regular • Anomalous • Each class has typical nodes. • Calculated through [Kleinburg99]

  11. Problem #4 • How to understand anomalous activities in a class?

  12. UPS Load Dock Human Data

  13. 1 month,9am-5pm, 5 days a week • 61 events, 10 key objects • 195 activities, 150 labelled • 7 major classes detected. (Table 1)

  14. Residential House Sensor Data • 5 months • 16 Strain gages • 16 event • every day is an activity

  15. Residential House Sensor Data • People seems to have different plans for different day. • 5 classes mined out. (Table 2)

  16. Kitchen Vision Data

  17. Whether activities like making salad, washing dishes will be detected? • Yes, with a proper n • 90% percent accuracy

  18. Anomalous Analysis Works • Discover some anomalous activities • Truck left with door open • Someone cleaning up the floor • …

  19. Activity-Class Characterization Presented by: Wei Pan For CS88/188

  20. Assume some activities in class c • 1-2-3-4-5 • 3-1-2-3-7 • 3-4-1-2-3-6-4-5 • 5-1-2-3-9-1 It seems 1-2-3 is very common in this class.

  21. Find a sequence of events s, so that s will have a certain prediction power in all activities of class c. Thus s will be a motif of class c. • Prediction power is analytically described as a bit-gain.

  22. Discussions and Questions

More Related