Activity Detection in Videos Riu Baring CIS 8590 Perception of Intelligent System Temple University Fall 2007.

Slides:



Advertisements
Similar presentations
Bayesian Belief Propagation
Advertisements

Probabilistic models Haixu Tang School of Informatics.
Change Detection C. Stauffer and W.E.L. Grimson, “Learning patterns of activity using real time tracking,” IEEE Trans. On PAMI, 22(8): , Aug 2000.
Part II – TIME SERIES ANALYSIS C5 ARIMA (Box-Jenkins) Models
1 Methods of Experimental Particle Physics Alexei Safonov Lecture #21.
Sampling distributions of alleles under models of neutral evolution.
EVSC 495/EVAT 795 Data Analysis & Climate Change Class hours: TuTh 2:00-3:15 pm Instructor: Michael E. Mann.
CHAPTER 16 MARKOV CHAIN MONTE CARLO
Anomaly Detection in the WIPER System using A Markov Modulated Poisson Distribution Ping Yan Tim Schoenharl Alec Pawling Greg Madey.
Chapter 7(7b): Statistical Applications in Traffic Engineering Chapter objectives: By the end of these chapters the student will be able to (We spend 3.
CS 589 Information Risk Management 6 February 2007.
Error Propagation. Uncertainty Uncertainty reflects the knowledge that a measured value is related to the mean. Probable error is the range from the mean.
Basics of Statistical Estimation. Learning Probabilities: Classical Approach Simplest case: Flipping a thumbtack tails heads True probability  is unknown.
Chapter 4 Probability Distributions
Student: Hsu-Yung Cheng Advisor: Jenq-Neng Hwang, Professor
Chapter Sampling Distributions and Hypothesis Testing.
CIS 2033 based on Dekking et al. A Modern Introduction to Probability and Statistics Instructor Longin Jan Latecki C12: The Poisson process.
Lecture Slides Elementary Statistics Twelfth Edition
8-2 Basics of Hypothesis Testing
Slide 1 Statistics Workshop Tutorial 7 Discrete Random Variables Binomial Distributions.
Bayesian Analysis for Extreme Events Pao-Shin Chu and Xin Zhao Department of Meteorology School of Ocean & Earth Science & Technology University of Hawaii-
Probabilistic Analysis of a Large-Scale Urban Traffic Sensor Data Set Jon Hutchins, Alexander Ihler, and Padhraic Smyth Department of Computer Science.
Jacinto C. Nascimento, Member, IEEE, and Jorge S. Marques
Modeling Count Data over Time Using Dynamic Bayesian Networks Jonathan Hutchins Advisors: Professor Ihler and Professor Smyth.
Elec471 Embedded Computer Systems Chapter 4, Probability and Statistics By Prof. Tim Johnson, PE Wentworth Institute of Technology Boston, MA Theory and.
WAC/ISSCI Automated Anomaly Detection Using Time-Variant Normal Profiling Jung-Yeop Kim, Utica College Rex E. Gantenbein, University of Wyoming.
©2003/04 Alessandro Bogliolo Background Information theory Probability theory Algorithms.
Lecture Slides Elementary Statistics Twelfth Edition
Inference for a Single Population Proportion (p).
Copyright © 2010, 2007, 2004 Pearson Education, Inc. Review and Preview This chapter combines the methods of descriptive statistics presented in.
Slide 1 Copyright © 2004 Pearson Education, Inc..
Section Copyright © 2014, 2012, 2010 Pearson Education, Inc. Lecture Slides Elementary Statistics Twelfth Edition and the Triola Statistics Series.
A Beginner’s Guide to Bayesian Modelling Peter England, PhD EMB GIRO 2002.
Measurement and Modeling of Packet Loss in the Internet Maya Yajnik.
Biostatistics Class 6 Hypothesis Testing: One-Sample Inference 2/29/2000.
1 An Interim Monitoring Approach for a Small Sample Size Incidence Density Problem By: Shane Rosanbalm Co-author: Dennis Wallace.
Copyright © 2010, 2007, 2004 Pearson Education, Inc. Section 5-2 Random Variables.
1 Let X represent a Binomial r.v as in (3-42). Then from (2-30) Since the binomial coefficient grows quite rapidly with n, it is difficult to compute (4-1)
A taste of statistics Normal error (Gaussian) distribution  most important in statistical analysis of data, describes the distribution of random observations.
ADAPTIVE EVENT DETECTION USING TIME-VARYING POISSON PROCESSES Kdd06 University of California, Irvine.
Image Modeling & Segmentation Aly Farag and Asem Ali Lecture #2.
- 1 - Bayesian inference of binomial problem Estimating a probability from binomial data –Objective is to estimate unknown proportion (or probability of.
A Model for Learning the Semantics of Pictures V. Lavrenko, R. Manmatha, J. Jeon Center for Intelligent Information Retrieval Computer Science Department,
The famous “sprinkler” example (J. Pearl, Probabilistic Reasoning in Intelligent Systems, 1988)
Learning to Detect Events with Markov-Modulated Poisson Processes Ihler, Hutchins and Smyth (2007)
The generalization of Bayes for continuous densities is that we have some density f(y|  ) where y and  are vectors of data and parameters with  being.
 Present by 陳群元.  Introduction  Previous work  Predicting motion patterns  Spatio-temporal transition distribution  Discerning pedestrians  Experimental.
Copyright © 2010, 2007, 2004 Pearson Education, Inc. All Rights Reserved. Section 5-1 Review and Preview.
A shared random effects transition model for longitudinal count data with informative missingness Jinhui Li Joint work with Yingnian Wu, Xiaowei Yang.
Education 793 Class Notes Inference and Hypothesis Testing Using the Normal Distribution 8 October 2003.
Copyright © 2010, 2007, 2004 Pearson Education, Inc. Lecture Slides Elementary Statistics Eleventh Edition and the Triola Statistics Series by.
Chapter 5 Probability Distributions 5-1 Overview 5-2 Random Variables 5-3 Binomial Probability Distributions 5-4 Mean, Variance and Standard Deviation.
Gibbs Sampling and Hidden Markov Models in the Event Detection Problem By Marc Sobel.
G. Cowan Lectures on Statistical Data Analysis Lecture 10 page 1 Statistical Data Analysis: Lecture 10 1Probability, Bayes’ theorem 2Random variables and.
Spatially Explicit Capture-recapture Models for Density Estimation 5.11 UF-2015.
Learning and Removing Cast Shadows through a Multidistribution Approach Nicolas Martel-Brisson, Andre Zaccarin IEEE TRANSACTIONS ON PATTERN ANALYSIS AND.
A Study on Speaker Adaptation of Continuous Density HMM Parameters By Chin-Hui Lee, Chih-Heng Lin, and Biing-Hwang Juang Presented by: 陳亮宇 1990 ICASSP/IEEE.
Probability Distributions ( 확률분포 ) Chapter 5. 2 모든 가능한 ( 확률 ) 변수의 값에 대해 확률을 할당하는 체계 X 가 1, 2, …, 6 의 값을 가진다면 이 6 개 변수 값에 확률을 할당하는 함수 Definition.
Modeling and Simulation CS 313
Probability Distributions: a review
Lecture Slides Elementary Statistics Eleventh Edition
Modeling and Simulation CS 313
A Non-Parametric Bayesian Method for Inferring Hidden Causes
Sequential Pattern Discovery under a Markov Assumption
Before-After Studies Part I
Particle Filters for Event Detection
Lecture Slides Elementary Statistics Twelfth Edition
Lecture Slides Elementary Statistics Twelfth Edition
Dept. of Electrical & Computer engineering
Presentation transcript:

Activity Detection in Videos Riu Baring CIS 8590 Perception of Intelligent System Temple University Fall 2007

Outline Background Related Work The Model Normal Count Event Count

Activity Detection Problems A process like e.g., traffic flow, crowd formation, or financial electronic transactions is unfolding in time. We can monitor and observe the flow frequencies at many fixed time points. Typically, there are many causes influencing changes in these frequencies.

Causes for Change Possible causes for change include: a) changes due to noise; i.e., those best modeled by e.g., a Gaussian error distribution. b) periodic changes; i.e., those expected to happen over periodic intervals. c) changes not due to either of the above: these are usually the changes we would like to detect.

An Example: Building Data 3 months of “people count” 30 minutes Calit2 UC Irvine Campus

Another Example: Traffic Data 6 months of estimated vehicle count Every 5 minutes Glendale on-ramp to 101N, Los Angeles

More Examples Detecting ‘Events’, which are not pre- planned, involving large numbers of people at a particular location. Detecting ‘Fraudulent transactions’. We observe a variety of electronic transactions at many time intervals. We would like to detect when the number of transactions is significantly different from what is expected.

Related Work Keogh et al. – KDD ‘02  Quantize real-valued time-series into finite set of symbols  Then use a Markov model to detect surprising patterns in the symbol sequence Guralnik and Srivastava – KDD ‘99  Iterative likelihood-based method for segmenting a time-series into homogeneous regions Salmenkivi and Mannila (2005)  Segmenting sets of low-level time-stamped events into time- periods of relatively constant intensity using a combination of Poisson models and Bayesian estimation methods Kleinberg – KDD ‘02  method based on an infinite automaton could be used to detect bursty events in text streams

Related Work All approaches share a common goal  detection of novel and unusual data points or segments in time-series. None focuses on detection of bursty events embedded in time series of counts that reflect the normal diurnal and calendar patterns of human activity. GOAL: To automatically detect the presence of unusual events in the observation sequence.

Background Markov-modulated Processes (Scott, 1998)  Analysis of Web Surfing Behavior (Scott and Smyth, 2005)  Telephone Network Fraud Detection (Scott, 2000) Ihler et al (KDD 2006) developed a framework for building a probabilistic model of time-varying counting process in which a superposition of both time-varying but regular (periodic) and aperiodic processes were observed.

Method I A Baseline Model  Where Threshold Adequate for:  Events interspersed in the data are sufficiently few  Events are sufficiently noticeable.

Chicken and Egg Method I Baseline Model Ideal Model

Method I False Positive, Persistence, and Duration Baseline ModelBaseline Model – Lower Threshold

Method II – Ideal Model Observed Count Normal Count (Unobserved) Event Count (Unobserved )

Normal Count Modeling Periodic Count Data d(t) = {1, …, 7} Sunday = 1, … h(t) = interval i.e. half-hour

Periodic Components Poisson Process Rate Day Effect Time of Day Effect

Event Count: The Process N E Events signify times during which there are higher frequencies which are not due to periodic or noise causes. We can model this by introducing a binary latent process z(t) and assuming z(t)=1 for such events and z(t)=0 if not. P(z(t)=1|z(t-1)=0)= 1-z 00 ; P(z(t)=0|z(t-1)=0)= z 00; P(z(t)=1|z(t-1)=1)= z 11 ; P(z(t)=0|z(t-1)=1)= 1-z 11 i.e., if there is no event at time t-1, the chance of an event at time t is 1-z 00 Modeling Rare Persistent Events

Graphical Model

Priors for event probabilities Beta distributions: priors for the z’s. and z 11 analogously. This characterizes the behavior of the underlying latent process. The hyperparameters a,b are designed to model that behavior.

Priors for event probabilities Recall that N 0 (t) (the non-event process) characterizes periodic and noise changes. The event process N E (t) characterizes other changes. N E (t) is 0 if z(t)=0 and Poisson with rate γ(t) if z(t)=1. So, if there is no event, N(t)=N 0 (t). If there is an event, the frequency due to periodic or noise changes is N(t)=N 0 (t)+N E (t) The rate γ (t) is itself gamma with parameters a E and b E. Hence (by conjugacy) it is marginally negative binomial (NB) with p=(b E /(1+b E ) and n=N(t).

Gibbs Sampling Gibbs sampling works by simulating each parameter/latent variable conditional on all the rest. The λ’s are parameters and the z’s,N’s are the latent variables. The resulting simulated values have an empirical distribution similar to the true posterior distribution. It works as a result of the fact that the joint distribution of parameters is determined by the set of all such conditional distributions.

Gibbs Sampling Given z(t)=0 and the remaining parameters, Put N 0 (t)=N(t) and N E (t)=0. If z(t)=1, simulate N E (t) as negative binomial with parameters, N(t) and b E /(1+b E ). Put N 0 (t)=N(t)-N E (t). To simulate z(t), define

More of Gibbs Sampling Then, if the previous state was 0, we get:

Gibbs Sampling (Continued) Having simulated z(t), we can simulate the parameters as follows: Where ‘N day ’ denotes the number of ‘day’ units in the data, ‘N hh ’ denotes the number of ‘hh’ periods in the data.

Result: Building Data

Result: Freeway Traffic Data

Result: 2600 frames

Discussion Poisson process (nonhomogeneous). Able to detect activity at the expected frame.

Future Work Histogram of direction implementation

References 1.A. Ihler, J. Hutchins, and P. Smyth, “Adaptive event detection with time-varying Poissons process,” KDD S. L. Scott and P. Smyth, “The Markov modulated Poisson process and Markov Poisson cascade with applications to web traffic data,” Bayesian Statistics, vol. 7, pp , S. L. Scott, “Detecting network intrusion using a Markov modulated nonhomogeneous Poisson process,” S. L. Scott, “Bayesian methods and extensions for the two state Markov modulated Poisson process,” Ph.D. dissertation, Harvard University, Dept. of Statistics, 1998.