SlideShare a Scribd company logo
CVPR 2012
Cross-view Activity Recognition using
            Hankelets
Binlong Li, Octavia I. Camps and Mario Sznaier
            Northeastern University




                                      Mobuddies
Dynamic Systems
 Dynamic systems have been recently used in a
  wide range of computer vision applications
 Given temporal sequence of observations
      (e.g. track coordinates) model temporal
  evolution as a function of low-dimensional state
  vector          that changes over time
 Simplest case – linear time invariant (LTI) system
  (w – noise)



 Practical limitation: given set of
  observations, triple      is not unique and is
Hankel Matrices
 Given a sequence of measurements
 , its block Hankel matrix is defined as:




 Columns correspond to overlapping
  subsequences of data
 Block anti-diagonals of the matrix are constant
 This structure encapsulates the dynamic
  information of the system
Initial condition invariance
 Linear time invariant system (LTI):


 In the absence of noice (w = 0):


 Then Hankel matrix is broken down to:




 Columns of Hankel matrix span the same
 subspace regardless of initial conditions
Autoregressive measurements
 Suppose the sequence of measurements is auto-
 regressive:

 Recall, that:




 Setting r = n in the above, we obtain:


 In other words, last column of Hankel matrix is a
 linear combination of other columns
Affine transformation invariance
 Suppose we have two Hankel matrices           and
  corresponding to a trajectory and its affine
  transformation. Auto-regressive property allows
  us to write:

 Suppose affine transformation is defined as
 Then, taking into account its linearity:


 In other words, sequences share the same
  autoregressor
 Recall, that
 Therefore, columns of two Hankel matrices span
Previous work
 B. Li et. al “Activity Recognition using Dynamic
    Subspace Angles”, CVPR 2011
   Considers initial condition invariance.
   Imagine that class of actions (e.g. “walk”) can be
    represented by a single dynamical system, and
    in-class variations are captured by different initial
    condition
   Then differentiating between two actions breaks
    down into determining whether columns of the
    two corresponding Hankel matrices lie in the
    same subspace
   Uses angles between subspaces as a measure of
Overview of the method
 Uses Dense trajectories to extract many short 15-
  frame tracklets.
 Builds Hankel matrix for each tracklet, capturing its
  velocity
 Employs BoF-like approach (BoHk)
 Does three experiments: single-view data, multiple
  view with knowledge transfer, multiple view without
  knowledge transfer
Hankelets
 Hankelet is a Hankel matrix for a short trajectory
 of 15 frames, formed by a sequence of
 normalized velocities:




 Normalize
  Hankelets:
Comparing Hankelets
 Introduce dissimilarity score between two
 Hankelets:

 Derivations show, that d ≈ 0 for Hankelets
 corresponding to noisy measurements of the
 same dynamical system
Building codebook: cluster center
 Modify the K-means algorithm for dissimilarity
  scores:
 Current Hankelet is assigned to a cluster whose
  “representative” has smallest dissimilarity with the
  current Hankelet
 Cluster’s “representative” is chosed as follows.
  Take random Hankelet within the class, find
  dissimilarities between the Hankelet and all other
  Hankelets in the cluster and compute their mean.
  The Hankelet with dissimilarity closest to the
  mean is selected as its “representative”
Building codebook: Gamma pdf
 The histogram of dissimilarities for a typical cluster in
  the dictionary of Hankelets:
 Represent each cluster
with its representative and
gamma pdf:



 Furthermore, each cluster
w has a prior probability
Bag of Hankelets (BoHk)
 Each activity video is represented with a
  histogram of labels from the dictionary of K
  Hankelets
 Cluster label is assigned using max probability:


   where       is cluster representative,    is
  cluster prior
 Finally, one-against-all non-linear SVM trained for
  activities recognition
Bi-Lingual Hankelets
 Bi-lingual Hankelets can be easily learned from
  unlabeled videos captured simultaneously from
  the different viewpoints by matching Hankelets
  across views (~80% are matched)
 Hankelets are matched using threshold on
  dissimilarity score, if their start times are the
  same (no spatial information)
Cross-view action recognition
 A labeled dataset is given, with Source and Target
 views
  Training
 Extract and match Bi-lingual Hankelets with
  dissimilarity score
 Build codebook of Bi-lingual Hankelets using the K-
  means
 Label Hankelets in Source data using max posterior
  probability
 Train one-against-all non-linear SVM using Source
  data
  Testing
Experiments
  Single-view
 KTH dataset: 95.89% avg.
  Cross-View with data transfer
 Use only Bi-lingual Hankelets
 IXMAS dataset: 56.4% avg. (45.5%
 improvement)
  Cross-View without data transfer
 Use all Hankelets (not only Bi-lingual)
 IXMAS dataset: 90.57% avg. (20.28%
 improvement)
Thank you!

More Related Content

What's hot

Mj upjs
Mj upjsMj upjs
Wavelet Signal Processing
Wavelet Signal ProcessingWavelet Signal Processing
Wavelet Signal Processing
HAmindavarLectures
 
Evaluating Island-based EAs on Unstable Networks with Complex Failure Patterns
Evaluating Island-based EAs on Unstable Networks with Complex Failure PatternsEvaluating Island-based EAs on Unstable Networks with Complex Failure Patterns
Evaluating Island-based EAs on Unstable Networks with Complex Failure Patterns
Rafael Nogueras
 
Packet Classification using Support Vector Machines with String Kernels
Packet Classification using Support Vector Machines with String KernelsPacket Classification using Support Vector Machines with String Kernels
Packet Classification using Support Vector Machines with String Kernels
IJERA Editor
 
Data Applied:Forecast
Data Applied:ForecastData Applied:Forecast
Data Applied:Forecast
DataminingTools Inc
 
Deep Recurrent Neural Network for Multi-target Filtering
Deep Recurrent Neural Network for Multi-target FilteringDeep Recurrent Neural Network for Multi-target Filtering
Deep Recurrent Neural Network for Multi-target Filtering
Mehryar (Mike) E., Ph.D.
 
Machine Learning Algorithms (Part 1)
Machine Learning Algorithms (Part 1)Machine Learning Algorithms (Part 1)
Machine Learning Algorithms (Part 1)
Zihui Li
 
A Study of the Performance of Self-* Memetic Algorithms on Heterogeneous Ephe...
A Study of the Performance of Self-* Memetic Algorithms on Heterogeneous Ephe...A Study of the Performance of Self-* Memetic Algorithms on Heterogeneous Ephe...
A Study of the Performance of Self-* Memetic Algorithms on Heterogeneous Ephe...
Rafael Nogueras
 
SEQUENTIAL CLUSTERING-BASED EVENT DETECTION FOR NONINTRUSIVE LOAD MONITORING
SEQUENTIAL CLUSTERING-BASED EVENT DETECTION FOR NONINTRUSIVE LOAD MONITORINGSEQUENTIAL CLUSTERING-BASED EVENT DETECTION FOR NONINTRUSIVE LOAD MONITORING
SEQUENTIAL CLUSTERING-BASED EVENT DETECTION FOR NONINTRUSIVE LOAD MONITORING
csandit
 
Grid based method & model based clustering method
Grid based method & model based clustering methodGrid based method & model based clustering method
Grid based method & model based clustering method
rajshreemuthiah
 

What's hot (11)

Mj upjs
Mj upjsMj upjs
Mj upjs
 
Wavelet Signal Processing
Wavelet Signal ProcessingWavelet Signal Processing
Wavelet Signal Processing
 
Evaluating Island-based EAs on Unstable Networks with Complex Failure Patterns
Evaluating Island-based EAs on Unstable Networks with Complex Failure PatternsEvaluating Island-based EAs on Unstable Networks with Complex Failure Patterns
Evaluating Island-based EAs on Unstable Networks with Complex Failure Patterns
 
Packet Classification using Support Vector Machines with String Kernels
Packet Classification using Support Vector Machines with String KernelsPacket Classification using Support Vector Machines with String Kernels
Packet Classification using Support Vector Machines with String Kernels
 
Data Applied:Forecast
Data Applied:ForecastData Applied:Forecast
Data Applied:Forecast
 
Deep Recurrent Neural Network for Multi-target Filtering
Deep Recurrent Neural Network for Multi-target FilteringDeep Recurrent Neural Network for Multi-target Filtering
Deep Recurrent Neural Network for Multi-target Filtering
 
Machine Learning Algorithms (Part 1)
Machine Learning Algorithms (Part 1)Machine Learning Algorithms (Part 1)
Machine Learning Algorithms (Part 1)
 
A Study of the Performance of Self-* Memetic Algorithms on Heterogeneous Ephe...
A Study of the Performance of Self-* Memetic Algorithms on Heterogeneous Ephe...A Study of the Performance of Self-* Memetic Algorithms on Heterogeneous Ephe...
A Study of the Performance of Self-* Memetic Algorithms on Heterogeneous Ephe...
 
Nn devs
Nn devsNn devs
Nn devs
 
SEQUENTIAL CLUSTERING-BASED EVENT DETECTION FOR NONINTRUSIVE LOAD MONITORING
SEQUENTIAL CLUSTERING-BASED EVENT DETECTION FOR NONINTRUSIVE LOAD MONITORINGSEQUENTIAL CLUSTERING-BASED EVENT DETECTION FOR NONINTRUSIVE LOAD MONITORING
SEQUENTIAL CLUSTERING-BASED EVENT DETECTION FOR NONINTRUSIVE LOAD MONITORING
 
Grid based method & model based clustering method
Grid based method & model based clustering methodGrid based method & model based clustering method
Grid based method & model based clustering method
 

Similar to Cross-view Activity Recognition using Hankelets

Termpaper ai
Termpaper aiTermpaper ai
Termpaper ai
Sumit Mishra
 
Special Aspects of Control for Quadrupedal Walking based on Biological Princi...
Special Aspects of Control for Quadrupedal Walking based on Biological Princi...Special Aspects of Control for Quadrupedal Walking based on Biological Princi...
Special Aspects of Control for Quadrupedal Walking based on Biological Princi...
Андрей Вуколов
 
Adaptive equalization
Adaptive equalizationAdaptive equalization
Adaptive equalization
Kamal Bhatt
 
TEST GENERATION FOR ANALOG AND MIXED-SIGNAL CIRCUITS USING HYBRID SYSTEM MODELS
TEST GENERATION FOR ANALOG AND MIXED-SIGNAL CIRCUITS USING HYBRID SYSTEM MODELSTEST GENERATION FOR ANALOG AND MIXED-SIGNAL CIRCUITS USING HYBRID SYSTEM MODELS
TEST GENERATION FOR ANALOG AND MIXED-SIGNAL CIRCUITS USING HYBRID SYSTEM MODELS
VLSICS Design
 
Test Generation for Analog and Mixed-Signal Circuits Using Hybrid System Mode...
Test Generation for Analog and Mixed-Signal Circuits Using Hybrid System Mode...Test Generation for Analog and Mixed-Signal Circuits Using Hybrid System Mode...
Test Generation for Analog and Mixed-Signal Circuits Using Hybrid System Mode...
VLSICS Design
 
Kane/DeAlbert dynamics for multibody system
Kane/DeAlbert dynamics for multibody system Kane/DeAlbert dynamics for multibody system
Kane/DeAlbert dynamics for multibody system
Tadele Belay
 
14 Machine Learning Single Layer Perceptron
14 Machine Learning Single Layer Perceptron14 Machine Learning Single Layer Perceptron
14 Machine Learning Single Layer Perceptron
Andres Mendez-Vazquez
 
Best data science course in pune. converted
Best data science course in pune. convertedBest data science course in pune. converted
Best data science course in pune. converted
sripadojwarumavilas
 
Best data science courses
Best data science coursesBest data science courses
Best data science courses
prathyusha1234
 
Data science institute in kolkata
Data science institute in kolkataData science institute in kolkata
Data science institute in kolkata
Bharath S
 
Python data science course
Python data science coursePython data science course
Python data science course
bhuvan8999
 
Machine learning
Machine learning Machine learning
Machine learning
mamatha08
 
Data science course in pune
Data science course in puneData science course in pune
Data science course in pune
Data Analytics Courses in Pune
 
Data science course in pune
Data science course in puneData science course in pune
Data science course in pune
prathyusha1234
 
Data science certification in pune
Data science certification in puneData science certification in pune
Data science certification in pune
prathyusha1234
 
Data science certification
Data science certificationData science certification
Data science certification
prathyusha1234
 
Machine learning certification in gurgaon
Machine learning certification in gurgaon Machine learning certification in gurgaon
Machine learning certification in gurgaon
TejaspathiLV
 
Data science training ang placements
Data science training ang placementsData science training ang placements
Data science training ang placements
bhuvan8999
 
Data science course
Data science courseData science course
Data science course
prathyusha1234
 

Similar to Cross-view Activity Recognition using Hankelets (20)

Termpaper ai
Termpaper aiTermpaper ai
Termpaper ai
 
Special Aspects of Control for Quadrupedal Walking based on Biological Princi...
Special Aspects of Control for Quadrupedal Walking based on Biological Princi...Special Aspects of Control for Quadrupedal Walking based on Biological Princi...
Special Aspects of Control for Quadrupedal Walking based on Biological Princi...
 
Adaptive equalization
Adaptive equalizationAdaptive equalization
Adaptive equalization
 
TEST GENERATION FOR ANALOG AND MIXED-SIGNAL CIRCUITS USING HYBRID SYSTEM MODELS
TEST GENERATION FOR ANALOG AND MIXED-SIGNAL CIRCUITS USING HYBRID SYSTEM MODELSTEST GENERATION FOR ANALOG AND MIXED-SIGNAL CIRCUITS USING HYBRID SYSTEM MODELS
TEST GENERATION FOR ANALOG AND MIXED-SIGNAL CIRCUITS USING HYBRID SYSTEM MODELS
 
Test Generation for Analog and Mixed-Signal Circuits Using Hybrid System Mode...
Test Generation for Analog and Mixed-Signal Circuits Using Hybrid System Mode...Test Generation for Analog and Mixed-Signal Circuits Using Hybrid System Mode...
Test Generation for Analog and Mixed-Signal Circuits Using Hybrid System Mode...
 
Kane/DeAlbert dynamics for multibody system
Kane/DeAlbert dynamics for multibody system Kane/DeAlbert dynamics for multibody system
Kane/DeAlbert dynamics for multibody system
 
14 Machine Learning Single Layer Perceptron
14 Machine Learning Single Layer Perceptron14 Machine Learning Single Layer Perceptron
14 Machine Learning Single Layer Perceptron
 
Unit 3
Unit 3Unit 3
Unit 3
 
Best data science course in pune. converted
Best data science course in pune. convertedBest data science course in pune. converted
Best data science course in pune. converted
 
Best data science courses
Best data science coursesBest data science courses
Best data science courses
 
Data science institute in kolkata
Data science institute in kolkataData science institute in kolkata
Data science institute in kolkata
 
Python data science course
Python data science coursePython data science course
Python data science course
 
Machine learning
Machine learning Machine learning
Machine learning
 
Data science course in pune
Data science course in puneData science course in pune
Data science course in pune
 
Data science course in pune
Data science course in puneData science course in pune
Data science course in pune
 
Data science certification in pune
Data science certification in puneData science certification in pune
Data science certification in pune
 
Data science certification
Data science certificationData science certification
Data science certification
 
Machine learning certification in gurgaon
Machine learning certification in gurgaon Machine learning certification in gurgaon
Machine learning certification in gurgaon
 
Data science training ang placements
Data science training ang placementsData science training ang placements
Data science training ang placements
 
Data science course
Data science courseData science course
Data science course
 

Recently uploaded

Unsubscribed: Combat Subscription Fatigue With a Membership Mentality by Head...
Unsubscribed: Combat Subscription Fatigue With a Membership Mentality by Head...Unsubscribed: Combat Subscription Fatigue With a Membership Mentality by Head...
Unsubscribed: Combat Subscription Fatigue With a Membership Mentality by Head...
Product School
 
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
Jeffrey Haguewood
 
Bits & Pixels using AI for Good.........
Bits & Pixels using AI for Good.........Bits & Pixels using AI for Good.........
Bits & Pixels using AI for Good.........
Alison B. Lowndes
 
Assuring Contact Center Experiences for Your Customers With ThousandEyes
Assuring Contact Center Experiences for Your Customers With ThousandEyesAssuring Contact Center Experiences for Your Customers With ThousandEyes
Assuring Contact Center Experiences for Your Customers With ThousandEyes
ThousandEyes
 
PCI PIN Basics Webinar from the Controlcase Team
PCI PIN Basics Webinar from the Controlcase TeamPCI PIN Basics Webinar from the Controlcase Team
PCI PIN Basics Webinar from the Controlcase Team
ControlCase
 
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
Product School
 
Key Trends Shaping the Future of Infrastructure.pdf
Key Trends Shaping the Future of Infrastructure.pdfKey Trends Shaping the Future of Infrastructure.pdf
Key Trends Shaping the Future of Infrastructure.pdf
Cheryl Hung
 
Epistemic Interaction - tuning interfaces to provide information for AI support
Epistemic Interaction - tuning interfaces to provide information for AI supportEpistemic Interaction - tuning interfaces to provide information for AI support
Epistemic Interaction - tuning interfaces to provide information for AI support
Alan Dix
 
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered QualitySoftware Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
Inflectra
 
Elevating Tactical DDD Patterns Through Object Calisthenics
Elevating Tactical DDD Patterns Through Object CalisthenicsElevating Tactical DDD Patterns Through Object Calisthenics
Elevating Tactical DDD Patterns Through Object Calisthenics
Dorra BARTAGUIZ
 
Empowering NextGen Mobility via Large Action Model Infrastructure (LAMI): pav...
Empowering NextGen Mobility via Large Action Model Infrastructure (LAMI): pav...Empowering NextGen Mobility via Large Action Model Infrastructure (LAMI): pav...
Empowering NextGen Mobility via Large Action Model Infrastructure (LAMI): pav...
Thierry Lestable
 
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
James Anderson
 
Designing Great Products: The Power of Design and Leadership by Chief Designe...
Designing Great Products: The Power of Design and Leadership by Chief Designe...Designing Great Products: The Power of Design and Leadership by Chief Designe...
Designing Great Products: The Power of Design and Leadership by Chief Designe...
Product School
 
FIDO Alliance Osaka Seminar: Passkeys and the Road Ahead.pdf
FIDO Alliance Osaka Seminar: Passkeys and the Road Ahead.pdfFIDO Alliance Osaka Seminar: Passkeys and the Road Ahead.pdf
FIDO Alliance Osaka Seminar: Passkeys and the Road Ahead.pdf
FIDO Alliance
 
AI for Every Business: Unlocking Your Product's Universal Potential by VP of ...
AI for Every Business: Unlocking Your Product's Universal Potential by VP of ...AI for Every Business: Unlocking Your Product's Universal Potential by VP of ...
AI for Every Business: Unlocking Your Product's Universal Potential by VP of ...
Product School
 
How world-class product teams are winning in the AI era by CEO and Founder, P...
How world-class product teams are winning in the AI era by CEO and Founder, P...How world-class product teams are winning in the AI era by CEO and Founder, P...
How world-class product teams are winning in the AI era by CEO and Founder, P...
Product School
 
Smart TV Buyer Insights Survey 2024 by 91mobiles.pdf
Smart TV Buyer Insights Survey 2024 by 91mobiles.pdfSmart TV Buyer Insights Survey 2024 by 91mobiles.pdf
Smart TV Buyer Insights Survey 2024 by 91mobiles.pdf
91mobiles
 
The Future of Platform Engineering
The Future of Platform EngineeringThe Future of Platform Engineering
The Future of Platform Engineering
Jemma Hussein Allen
 
From Daily Decisions to Bottom Line: Connecting Product Work to Revenue by VP...
From Daily Decisions to Bottom Line: Connecting Product Work to Revenue by VP...From Daily Decisions to Bottom Line: Connecting Product Work to Revenue by VP...
From Daily Decisions to Bottom Line: Connecting Product Work to Revenue by VP...
Product School
 
Neuro-symbolic is not enough, we need neuro-*semantic*
Neuro-symbolic is not enough, we need neuro-*semantic*Neuro-symbolic is not enough, we need neuro-*semantic*
Neuro-symbolic is not enough, we need neuro-*semantic*
Frank van Harmelen
 

Recently uploaded (20)

Unsubscribed: Combat Subscription Fatigue With a Membership Mentality by Head...
Unsubscribed: Combat Subscription Fatigue With a Membership Mentality by Head...Unsubscribed: Combat Subscription Fatigue With a Membership Mentality by Head...
Unsubscribed: Combat Subscription Fatigue With a Membership Mentality by Head...
 
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
 
Bits & Pixels using AI for Good.........
Bits & Pixels using AI for Good.........Bits & Pixels using AI for Good.........
Bits & Pixels using AI for Good.........
 
Assuring Contact Center Experiences for Your Customers With ThousandEyes
Assuring Contact Center Experiences for Your Customers With ThousandEyesAssuring Contact Center Experiences for Your Customers With ThousandEyes
Assuring Contact Center Experiences for Your Customers With ThousandEyes
 
PCI PIN Basics Webinar from the Controlcase Team
PCI PIN Basics Webinar from the Controlcase TeamPCI PIN Basics Webinar from the Controlcase Team
PCI PIN Basics Webinar from the Controlcase Team
 
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
 
Key Trends Shaping the Future of Infrastructure.pdf
Key Trends Shaping the Future of Infrastructure.pdfKey Trends Shaping the Future of Infrastructure.pdf
Key Trends Shaping the Future of Infrastructure.pdf
 
Epistemic Interaction - tuning interfaces to provide information for AI support
Epistemic Interaction - tuning interfaces to provide information for AI supportEpistemic Interaction - tuning interfaces to provide information for AI support
Epistemic Interaction - tuning interfaces to provide information for AI support
 
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered QualitySoftware Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
 
Elevating Tactical DDD Patterns Through Object Calisthenics
Elevating Tactical DDD Patterns Through Object CalisthenicsElevating Tactical DDD Patterns Through Object Calisthenics
Elevating Tactical DDD Patterns Through Object Calisthenics
 
Empowering NextGen Mobility via Large Action Model Infrastructure (LAMI): pav...
Empowering NextGen Mobility via Large Action Model Infrastructure (LAMI): pav...Empowering NextGen Mobility via Large Action Model Infrastructure (LAMI): pav...
Empowering NextGen Mobility via Large Action Model Infrastructure (LAMI): pav...
 
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
 
Designing Great Products: The Power of Design and Leadership by Chief Designe...
Designing Great Products: The Power of Design and Leadership by Chief Designe...Designing Great Products: The Power of Design and Leadership by Chief Designe...
Designing Great Products: The Power of Design and Leadership by Chief Designe...
 
FIDO Alliance Osaka Seminar: Passkeys and the Road Ahead.pdf
FIDO Alliance Osaka Seminar: Passkeys and the Road Ahead.pdfFIDO Alliance Osaka Seminar: Passkeys and the Road Ahead.pdf
FIDO Alliance Osaka Seminar: Passkeys and the Road Ahead.pdf
 
AI for Every Business: Unlocking Your Product's Universal Potential by VP of ...
AI for Every Business: Unlocking Your Product's Universal Potential by VP of ...AI for Every Business: Unlocking Your Product's Universal Potential by VP of ...
AI for Every Business: Unlocking Your Product's Universal Potential by VP of ...
 
How world-class product teams are winning in the AI era by CEO and Founder, P...
How world-class product teams are winning in the AI era by CEO and Founder, P...How world-class product teams are winning in the AI era by CEO and Founder, P...
How world-class product teams are winning in the AI era by CEO and Founder, P...
 
Smart TV Buyer Insights Survey 2024 by 91mobiles.pdf
Smart TV Buyer Insights Survey 2024 by 91mobiles.pdfSmart TV Buyer Insights Survey 2024 by 91mobiles.pdf
Smart TV Buyer Insights Survey 2024 by 91mobiles.pdf
 
The Future of Platform Engineering
The Future of Platform EngineeringThe Future of Platform Engineering
The Future of Platform Engineering
 
From Daily Decisions to Bottom Line: Connecting Product Work to Revenue by VP...
From Daily Decisions to Bottom Line: Connecting Product Work to Revenue by VP...From Daily Decisions to Bottom Line: Connecting Product Work to Revenue by VP...
From Daily Decisions to Bottom Line: Connecting Product Work to Revenue by VP...
 
Neuro-symbolic is not enough, we need neuro-*semantic*
Neuro-symbolic is not enough, we need neuro-*semantic*Neuro-symbolic is not enough, we need neuro-*semantic*
Neuro-symbolic is not enough, we need neuro-*semantic*
 

Cross-view Activity Recognition using Hankelets

  • 1. CVPR 2012 Cross-view Activity Recognition using Hankelets Binlong Li, Octavia I. Camps and Mario Sznaier Northeastern University Mobuddies
  • 2. Dynamic Systems  Dynamic systems have been recently used in a wide range of computer vision applications  Given temporal sequence of observations (e.g. track coordinates) model temporal evolution as a function of low-dimensional state vector that changes over time  Simplest case – linear time invariant (LTI) system (w – noise)  Practical limitation: given set of observations, triple is not unique and is
  • 3. Hankel Matrices  Given a sequence of measurements , its block Hankel matrix is defined as:  Columns correspond to overlapping subsequences of data  Block anti-diagonals of the matrix are constant  This structure encapsulates the dynamic information of the system
  • 4. Initial condition invariance  Linear time invariant system (LTI):  In the absence of noice (w = 0):  Then Hankel matrix is broken down to:  Columns of Hankel matrix span the same subspace regardless of initial conditions
  • 5. Autoregressive measurements  Suppose the sequence of measurements is auto- regressive:  Recall, that:  Setting r = n in the above, we obtain:  In other words, last column of Hankel matrix is a linear combination of other columns
  • 6. Affine transformation invariance  Suppose we have two Hankel matrices and corresponding to a trajectory and its affine transformation. Auto-regressive property allows us to write:  Suppose affine transformation is defined as  Then, taking into account its linearity:  In other words, sequences share the same autoregressor  Recall, that  Therefore, columns of two Hankel matrices span
  • 7. Previous work  B. Li et. al “Activity Recognition using Dynamic Subspace Angles”, CVPR 2011  Considers initial condition invariance.  Imagine that class of actions (e.g. “walk”) can be represented by a single dynamical system, and in-class variations are captured by different initial condition  Then differentiating between two actions breaks down into determining whether columns of the two corresponding Hankel matrices lie in the same subspace  Uses angles between subspaces as a measure of
  • 8. Overview of the method  Uses Dense trajectories to extract many short 15- frame tracklets.  Builds Hankel matrix for each tracklet, capturing its velocity  Employs BoF-like approach (BoHk)  Does three experiments: single-view data, multiple view with knowledge transfer, multiple view without knowledge transfer
  • 9. Hankelets  Hankelet is a Hankel matrix for a short trajectory of 15 frames, formed by a sequence of normalized velocities:  Normalize Hankelets:
  • 10. Comparing Hankelets  Introduce dissimilarity score between two Hankelets:  Derivations show, that d ≈ 0 for Hankelets corresponding to noisy measurements of the same dynamical system
  • 11. Building codebook: cluster center  Modify the K-means algorithm for dissimilarity scores:  Current Hankelet is assigned to a cluster whose “representative” has smallest dissimilarity with the current Hankelet  Cluster’s “representative” is chosed as follows. Take random Hankelet within the class, find dissimilarities between the Hankelet and all other Hankelets in the cluster and compute their mean. The Hankelet with dissimilarity closest to the mean is selected as its “representative”
  • 12. Building codebook: Gamma pdf  The histogram of dissimilarities for a typical cluster in the dictionary of Hankelets:  Represent each cluster with its representative and gamma pdf:  Furthermore, each cluster w has a prior probability
  • 13. Bag of Hankelets (BoHk)  Each activity video is represented with a histogram of labels from the dictionary of K Hankelets  Cluster label is assigned using max probability: where is cluster representative, is cluster prior  Finally, one-against-all non-linear SVM trained for activities recognition
  • 14. Bi-Lingual Hankelets  Bi-lingual Hankelets can be easily learned from unlabeled videos captured simultaneously from the different viewpoints by matching Hankelets across views (~80% are matched)  Hankelets are matched using threshold on dissimilarity score, if their start times are the same (no spatial information)
  • 15.
  • 16. Cross-view action recognition  A labeled dataset is given, with Source and Target views Training  Extract and match Bi-lingual Hankelets with dissimilarity score  Build codebook of Bi-lingual Hankelets using the K- means  Label Hankelets in Source data using max posterior probability  Train one-against-all non-linear SVM using Source data Testing
  • 17. Experiments Single-view  KTH dataset: 95.89% avg. Cross-View with data transfer  Use only Bi-lingual Hankelets  IXMAS dataset: 56.4% avg. (45.5% improvement) Cross-View without data transfer  Use all Hankelets (not only Bi-lingual)  IXMAS dataset: 90.57% avg. (20.28% improvement)