Equine Gait Analysis and Visualization Methods Dr. Marjorie Skubic Samer Arafat Justin Satterley Computer Engineering & Computer Science Dr. Kevin Keegan.

Slides:



Advertisements
Similar presentations
[1] AN ANALYSIS OF DIGITAL WATERMARKING IN FREQUENCY DOMAIN.
Advertisements

Noise & Data Reduction. Paired Sample t Test Data Transformation - Overview From Covariance Matrix to PCA and Dimension Reduction Fourier Analysis - Spectrum.
1 VLDB 2006, Seoul Mapping a Moving Landscape by Mining Mountains of Logs Automated Generation of a Dependency Model for HUG’s Clinical System Mirko Steinle,
Principal Component Analysis Based on L1-Norm Maximization Nojun Kwak IEEE Transactions on Pattern Analysis and Machine Intelligence, 2008.
An Approach to ECG Delineation using Wavelet Analysis and Hidden Markov Models Maarten Vaessen (FdAW/Master Operations Research) Iwan de Jong (IDEE/MI)
Tim Smyth and Jamie Shutler Assessment of analysis and forecast skill Assessment using satellite data.
Introduction and Overview Dr Mohamed A. El-Gebeily Department of Mathematical Sciences KFUPM
Clinical Assessment of Locomotion GaitTrak Gait Analysis - Why ? Gait Analysis - Why ? Applications for Gait Analysis Applications for Gait Analysis Gait.
Introduction The aim the project is to analyse non real time EEG (Electroencephalogram) signal using different mathematical models in Matlab to predict.
Pitch Recognition with Wavelets Final Presentation by Stephen Geiger.
Accelerometer-based Transportation Mode Detection on Smartphones
WRSTA, 13 August, 2006 Rough Sets in Hybrid Intelligent Systems For Breast Cancer Detection By Aboul Ella Hassanien Cairo University, Faculty of Computer.
Content-Based Classification, Search & Retrieval of Audio Erling Wold, Thom Blum, Douglas Keislar, James Wheaton Presented By: Adelle C. Knight.
Advanced Computer Graphics (Fall 2010) CS 283, Lecture 24: Motion Capture Ravi Ramamoorthi Most slides courtesy.
TRADING OFF PREDICTION ACCURACY AND POWER CONSUMPTION FOR CONTEXT- AWARE WEARABLE COMPUTING Presented By: Jeff Khoshgozaran.
Lecture 5: Learning models using EM
Reengineering Classification at the USPTO Marti Hearst, Chief IT Strategist, USPTO PIUG Conference May 4, 2010.
Speaker Adaptation for Vowel Classification
Multi-Scale Analysis for Network Traffic Prediction and Anomaly Detection Ling Huang Joint work with Anthony Joseph and Nina Taft January, 2005.
AI – CS364 Hybrid Intelligent Systems Overview of Hybrid Intelligent Systems 07 th November 2005 Dr Bogdan L. Vrusias
Optimal Adaptation for Statistical Classifiers Xiao Li.
Multimedia Security Digital Video Watermarking Supervised by Prof. LYU, Rung Tsong Michael Presented by Chan Pik Wah, Pat Nov 20, 2002 Department of Computer.
Pattern Recognition. Introduction. Definitions.. Recognition process. Recognition process relates input signal to the stored concepts about the object.
Object Detection Using the Statistics of Parts Henry Schneiderman Takeo Kanade Presented by : Sameer Shirdhonkar December 11, 2003.
Hub Queue Size Analyzer Implementing Neural Networks in practice.
Handwritten Character Recognition using Hidden Markov Models Quantifying the marginal benefit of exploiting correlations between adjacent characters and.
Unit 3a Industrial Control Systems
Extracting Places and Activities from GPS Traces Using Hierarchical Conditional Random Fields Yong-Joong Kim Dept. of Computer Science Yonsei.
Approximating the Algebraic Solution of Systems of Interval Linear Equations with Use of Neural Networks Nguyen Hoang Viet Michal Kleiber Institute of.
Electrical and Computer Systems Engineering Postgraduate Student Research Forum 2001 WAVELET ANALYSIS FOR CONDITION MONITORING OF CIRCUIT BREAKERS Author:
© Copyright 2004 ECE, UM-Rolla. All rights reserved A Brief Overview of Neural Networks By Rohit Dua, Samuel A. Mulder, Steve E. Watkins, and Donald C.
Texture analysis Team 5 Alexandra Bulgaru Justyna Jastrzebska Ulrich Leischner Vjekoslav Levacic Güray Tonguç.
Compiled By: Raj G Tiwari.  A pattern is an object, process or event that can be given a name.  A pattern class (or category) is a set of patterns sharing.
Texture. Texture is an innate property of all surfaces (clouds, trees, bricks, hair etc…). It refers to visual patterns of homogeneity and does not result.
DIGITAL WATERMARKING SRINIVAS KHARSADA PATNAIK [1] AN ANALYSIS OF DIGITAL WATERMARKING IN FREQUENCY DOMAIN Presented by SRINIVAS KHARSADA PATNAIK ROLL.
Time Series Data Analysis - I Yaji Sripada. Dept. of Computing Science, University of Aberdeen2 In this lecture you learn What are Time Series? How to.
R. Ray and K. Chen, department of Computer Science engineering  Abstract The proposed approach is a distortion-specific blind image quality assessment.
MUMT611: Music Information Acquisition, Preservation, and Retrieval Presentation on Timbre Similarity Alexandre Savard March 2006.
Overview of Part I, CMSC5707 Advanced Topics in Artificial Intelligence KH Wong (6 weeks) Audio signal processing – Signals in time & frequency domains.
A Study of Residue Correlation within Protein Sequences and its Application to Sequence Classification Christopher Hemmerich Advisor: Dr. Sun Kim.
Indian Institute of Information Technology and Management Gwalior24/12/2008 DR. ANUPAM SHUKLA DR. RITU TIWARI HEMANT KUMAR MEENA RAHUL KALA Speaker Identification.
Content-Based Image Retrieval Using Fuzzy Cognition Concepts Presented by Tienwei Tsai Department of Computer Science and Engineering Tatung University.
CCN COMPLEX COMPUTING NETWORKS1 This research has been supported in part by European Commission FP6 IYTE-Wireless Project (Contract No: )
The Wavelet Tutorial: Part2 Dr. Charturong Tantibundhit.
Data Mining and Decision Support
CSC321 Lecture 5 Applying backpropagation to shape recognition Geoffrey Hinton.
EEG processing based on IFAST system and Artificial Neural Networks for early detection of Alzheimer’s disease.
Learning to Rank: From Pairwise Approach to Listwise Approach Authors: Zhe Cao, Tao Qin, Tie-Yan Liu, Ming-Feng Tsai, and Hang Li Presenter: Davidson Date:
In The Name of God The Compassionate The Merciful.
Computational Intelligence: Methods and Applications Lecture 26 Density estimation, Expectation Maximization. Włodzisław Duch Dept. of Informatics, UMK.
3D Motion Classification Partial Image Retrieval and Download Multimedia Project Multimedia and Network Lab, Department of Computer Science.
An E-Textiles. Virginia Tech e-Textiles Group Design of an e-textile computer architecture – Networking – Fault tolerance – Power aware – Programming.
Chapter 11 – Neural Nets © Galit Shmueli and Peter Bruce 2010 Data Mining for Business Intelligence Shmueli, Patel & Bruce.
Deep Learning Amin Sobhani.
3D Motion Classification Partial Image Retrieval and Download
Gender Classification Using Scaled Conjugate Gradient Back Propagation
Supervised Time Series Pattern Discovery through Local Importance
Automatic Sleep Stage Classification using a Neural Network Algorithm
Vijay Srinivasan Thomas Phan
Vincent Granville, Ph.D. Co-Founder, DSC
Audio Content Description
Multi-resolution analysis
The El Nino Time Series A classically difficult problem because of nested features and different time scales.
Enhancing Diagnostic Quality of ECG in Mobile Environment
Audio and Speech Computers & New Media.
Synthesis of Motion from Simple Animations
Model Enhanced Classification of Serious Adverse Events
Data Preprocessing Copyright, 1996 © Dale Carnegie & Associates, Inc.
Data Preprocessing Copyright, 1996 © Dale Carnegie & Associates, Inc.
The El Nino Time Series A classically difficult problem because of nested features and different time scales.
Presentation transcript:

Equine Gait Analysis and Visualization Methods Dr. Marjorie Skubic Samer Arafat Justin Satterley Computer Engineering & Computer Science Dr. Kevin Keegan Veterinary Medicine & Surgery

Motion capture Raw data Transformed data for analysis Classification right lame left lame sound Overview Animation for visualization Database Pre-process and store

Motion capture Raw data Transformed data for analysis Animation for visualization Database Pre-process and store Classification Can this be applied to human motion?

Animation for visualization Database Pre-process and store Motion capture Raw data Transformed data for analysis Classification right lame left lame sound Analysis and Classification

Gait Analysis Cycle Measurement of walking biomechanics. Computation of temporal parameters, body kinematics, or EMG signals. Identification, assessment, and characterization of abnormal gait. Recommendations for treatment alternatives. Periodic analysis post intervention measures improvement.

Difficult Problem Wealth of information. Complexity of motion. Uncertainty about gait data quality. Mild lameness problem difficulty. Formulating a generalized method

Examples

Computerized Analysis Provides objective evaluation of interrelationships between observed body parts Signal Processing techniques: –Fourier Preprocessing Fixed frequency window not suited for short duration pulsation Few harmonics represent signal details Produces no time domain localization –Discrete Wavelet Preprocessing Limited window (scale) widths, at 1,2,4,8,16,32,… Limited on time localization. –Continuous Wavelet Preprocessing data collection preprocessing classification

Fourier Preprocessing Holzreiter1993 and Lakany1997 showed good results for the 2-class problem: sound vs. lame gait. Fourier Analysis: although localized in frequency domain, fixed frequency window not suited for short duration pulsation; few harmonics represent signal details; produces no time domain localization. Lakany2000 concluded that wavelet transform has the advantage of extracting local or global features.

Discrete Wavelet Preprocessing Marguitu1997, Verdini2000, and Sekine2000 showed good results for the 2-class problem. DWT has limited window (scale) widths, at 1,2,4,8,16,32,… DWT is limited on time localization.

Continuous Wavelet Preprocessing Lakany 2000 showed good results for the 2-class gait problem: sound vs. lame. CWT has temporal localization. Has flexible window sizes. Is translation invariant. Can be used to extract generic features: local and global signal characteristics.

CWT Coefficients CWT may be thought of as a rough measure of similarity between wavelet and signal segment. Need to select wavelet most similar to signal characteristics. Example Wavelets

Wavelet Selection Standard method is to: 1. Do a visual inspection of signal characteristics and available wavelets. 2. Select a wavelet that “looks” similar to dominant signal characteristics. Examples: Aminian 2002, Ismail1 998, Lakany Method is subjective, time-consuming, manual, and imprecise (most similar, or best, wavelet might not get selected).

Automatic Wavelet Selection Need a method that searches for a wavelet that is maximally similar to signal characteristics. Analyze information content of transformed signals. System’s self-information is related to uncertainty [Shannon 1949]. Maximum entropy yields highest self-information.

Uncertainty Types Complex information systems exhibit several types of uncertainty [Pal2000], [Yager2000]. Include - Probabilistic: uncertainty due to randomness. - Fuzzy: measures average ambiguity in fuzzy sets. - Non-specific: ambiguity in specifying exact solution.

Combined Uncertainty Shannon 1949 introduced maximum entropy, which is a probabilistic uncertainty measure. We explore a generalization that includes fuzzy and probabilistic uncertainties. Fuzzy and probabilistic uncertainties are combined together in order to compute maximum uncertainty. Better models system self-information.

Best Wavelet Selection Select an initial set of scales: 16,32,52,64. For each scale value, For each Horse data set, For each available wavelet Compute CWT Compute Coefficient’s Uncertainty Horse’s B.W. has Maximum Uncertainty Best Wavelet is selected most often by Horses.

Best Transformation Analysis

Time Sequence Process TS process combines together 3, 5, or 7 adjacent transformed signal data points. Captures intra-signal variation over time. Composition captures temporal trend. TS points form feature vectors that are input to NN. Helps NN combine together multiple signals in order to capture their temporal correlations.

Forming Feature Vectors for a Neural Network Classifier

Gait Classification Experiments Navicular data set: used 8 horses/class. Used BP neural nets for training with conjugate gradient algorithm. Used 6-fold for training, 2-fold for testing. Correct classification percentage (CCP) computed 8 experiments make 1 round. 7 rounds total. Median CCP is recorded.

Navicular Set Results

Induced-Lameness Set Special shoes attached to reasonably sound horses in order to induce a level of lameness on a certain side. Two level sets: mild and severe. Severe used here. Navicular set trained NN used to pick good horses. 9 horses determined to be good on all 3 classes. BWS with Combined Uncertainty picked best wavelets. 95% CCP recorded.

Fetlock, Elbow, & Carpus 2 points suggested by medical practitioners to pick side of lameness: poll and foot. Multiple features extracted per signal. Single features scored low CCP. Multiple features improved performance (83% CCP). Poll and foot needed only one feature. Poll + one leg point can pick side of lameness. Foot is best point.

Small Feature Extraction Used BWS with CU to extract foot’s small feature. Computed 87% CCP. Information in small features Zoom-in on desired features. Avoid scales < 6

Intermediate Conclusions BWS algorithm may be used to extract gait signal characteristics. TS process captures intra-signal trend changes. Combined Uncertainty better models system’s self- information, compared to Prob. or Fuzzy Uncertainty. BWS using CU algorithm automatically selects wavelets that are most similar to generic periodic signals. Shannon’s maximum entropy may be generalized to maximum combined uncertainty. Poll + 1 leg signal enough to characterize lameness, with the foot being the best leg point.

Future Plan Experiment with new key points in induced- lameness data set. Investigate other uncertainty types, like non- specificity. Evaluate methods using synthetic data. Evaluate induced-lameness data using NN trained with induced-lameness data and tested on navicular data set.

Animation for visualization Database Pre-process and store Motion capture Raw data Transformed data for analysis Classification right lame left lame sound Visualization Methods

RideHP

RideHP Raw Data (Pitch) Time Angular Velocity Integrated Data (Pitch) Time Position

RideHP Integrated Data (Pitch) Time Position Adjusted Integrated Data (Pitch) Time Position

RideHP Slowed (75%) Side View

Motion capture Raw data Transformed data for analysis Animation for visualization Database Pre-process and store Classification Can this be applied to human motion?

Possible Application to Human Motion Monitoring treatments for injuries and disabilities –Is the treatment working? Monitoring the elderly –Detect mobility deterioration –Start preventative exercise Monitoring movement for sports performance

Questions? Contact information: Web: