Scribd Logo
Upload

Welcome to Scribd!

  • 18 views

    Session No: CO2-1 Session Topic: Motion Analysis: Digital Video Processing

    Uploaded by

    kamal nadh
    The document discusses motion analysis and the Gaussian mixture model (GMM). It begins with an overview of motion analysis, including how it works by comparing frames to detect movement, and its uses in fields like manufacturing, healthcare, and autonomous vehicles. It then explains that GMM is an unsupervised learning technique used for clustering data points. GMM models the data as a probability density mixture of multiple Gaussian distributions. The Expectation-Maximization algorithm is used to estimate the parameters for GMM by filling in any missing data points.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd

    Session No: CO2-1 Session Topic: Motion Analysis: Digital Video Processing

    Uploaded by

    kamal nadh
    18 views29 pages
    The document discusses motion analysis and the Gaussian mixture model (GMM). It begins with an overview of motion analysis, including how it works by comparing frames to detect movement, and its uses in fields like manufacturing, healthcare, and autonomous vehicles. It then explains that GMM is an unsupervised learning technique used for clustering data points. GMM models the data as a probability density mixture of multiple Gaussian distributions. The Expectation-Maximization algorithm is used to estimate the parameters for GMM by filling in any missing data points.

    Original Title

    DVP_19cs3278_CO_2_SESSION_1

    Copyright

    © © All Rights Reserved

    Available Formats

    PPTX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document discusses motion analysis and the Gaussian mixture model (GMM). It begins with an overview of motion analysis, including how it works by comparing frames to detect movement, and its uses in fields like manufacturing, healthcare, and autonomous vehicles. It then explains that GMM is an unsupervised learning technique used for clustering data points. GMM models the data as a probability density mixture of multiple Gaussian distributions. The Expectation-Maximization algorithm is used to estimate the parameters for GMM by filling in any missing data points.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    18 views29 pages

    Session No: CO2-1 Session Topic: Motion Analysis: Digital Video Processing

    Uploaded by

    kamal nadh
    The document discusses motion analysis and the Gaussian mixture model (GMM). It begins with an overview of motion analysis, including how it works by comparing frames to detect movement, and its uses in fields like manufacturing, healthcare, and autonomous vehicles. It then explains that GMM is an unsupervised learning technique used for clustering data points. GMM models the data as a probability density mixture of multiple Gaussian distributions. The Expectation-Maximization algorithm is used to estimate the parameters for GMM by filling in any missing data points.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    of 29

    Session No: CO2-1

    Session Topic: Motion Analysis

    DIGITAL VIDEO PROCESSING


    (Course code: 19cs3278)
    Prepared & Presented by: Dr. Lambodar Jena
    Session Objective
    To understand the
    • Motion analysis
    • GMM model
    Poll Question-01
    • A video consists of a sequence of
    A. Frames
    B. Signals
    C. Packets
    D. Slots
    Key Concepts
    • Introduction to Motion Analysis
    • How motion analysis works?
    • Uses of motion analysis
    Motion Analysis

    • Motion analysis is a measuring technique used in computer vision,


    image processing and high-speed photography applications to detect
    movement.
    Objectives of motion analysis

    The objectives of motion analysis are


    to detect motion within an image
    track an object’s motion over time
    group objects that move together
    identify the direction of motion
    Specific techniques
    Specific techniques for implementing motion or movement analysis
    include
    electromyography,
    background segmentation and
    differential equation models
    How motion analysis works

    • The basic function of motion analysis is to compare two or more


    consecutive images captured by sensors or cameras to return
    information on the apparent motion in the images.
    • This is usually done by programming the recording device to produce 
    binary images based on movement.
    • All of the image points, or pixels, that correspond with motion are set
    to a value of 1 while stationary pixels are set to 0.
    How motion analysis works cont…

    • The resulting image can be processed even further to remove noise,


    label objects and group neighboring values of 1 into a singular object.
    • The data produced by motion analysis tools often correlates to a
    specific image at a specific point in time based on its position in the
    sequence.
    • Therefore, the motion capture data is time-dependent, which is a
    crucial component in most tracking applications.
    Uses of motion analysis
    • Motion analysis is used in a variety of fields and applications,
    including:
    1. Manufacturing- Motion analysis can be applied to the
    manufacturing process through software that monitors and
    analyzes supply chains for inefficiencies or malfunctions. Similarly,
    motion analysis can be used by manufacturers to conduct product
    safety, collision or efficiency tests.
    2. Video surveillance- Human activity recognition through motion
    analysis is commonly used for security monitoring and surveillance
    purposes.
    Uses of motion analysis cont…
    3. Healthcare and physical therapy- Motion analysis can help
    healthcare providers track muscle activity, perform gait analysis and
    diagnose potential mobility issues. Treatment for patients with
    cerebral palsy, spina bifida, muscular dystrophy and joint issues can
    involve regular motion analysis tests in a laboratory.
    4. Autonomous vehicles- Motion analysis systems can be used in 
    self-driving cars to aid with traffic navigation and obstruction
    identification.
    5. Biological sciences- Specific motion analysis software can be used
    to count and track tiny particles such as bacteria and viruses.
    Why Gaussian Mixture Model (GMM) in
    Image Processing?
    • In Machine Learning, there are two areas which are- Supervised
    learning and Unsupervised Learning.
    • The difference between both is the approach used to solve the
    problem statement and the data used in the approach.
    • There is a term in Unsupervised Learning called Clustering in which
    we find the clusters of data points with some common characteristics.
    Example:

    Fig. : Datapoints
    • Now clustering means to find the set of points that are close together
    as compared to some other data points.

    Here there are two sets of


    clusters or set of data points
    that are close together. The
    two sets are represented in
    blue and red.

    Fig. : Clusters
    Why Gaussian Mixture Model cont…
    • Here in the above image, there is one more notation for the centroids
    of both clusters and act as the parameters that identify each one of
    the clusters.
    • There are multiple methods or approaches that are used to do the
    clustering. They are-
    K-means Clustering
    Gaussian Mixture Model
    Hierarchical Clustering
    etc….
    Why Gaussian Mixture Model cont…
    • K-means is quite a popular clustering algorithm that updates the
    parameters of each cluster by an iterative approach. Basically, it
    calculates the centroid (means) of each cluster and then calculates
    their distance to each of the data points. This process is repeated until
    some criteria are fulfilled. 
    • K-means is a hard clustering method which means that it associates
    each point to one and only cluster. The limitation of this method is
    that there is no probability that tells us how much a data point is
    associated with the cluster.
    • This is when GMM (Gaussian Mixture Model) comes to the picture.
    • Let’s recall types of clustering methods:
    • Hard clustering: clusters do not overlap (element either belongs to
    cluster or it does not) — e.g. K-means, K-Medoid.
    • Soft clustering: clusters may overlap (strength of association between
    clusters and instances) — e.g. mixture model using Expectation-
    Maximization algorithm.
    GMM (Gaussian Mixture Model)
    • The core idea of this model is that it tries to model the dataset in the
    mixture of multiple Gaussian mixtures.
    What is Gaussian Mixture ?
    • Gaussian Mixture is a function that includes multiple Gaussians equal
    to the total number of clusters formed.
    • Each Gaussian in the mixture carries some parameters which are- 
    A mean, that defines the center.
    A covariance, that defines the width.
    A probability.
    Fig. : GMM clusters
    • Here it can be seen that there are three clusters that mean three
    Gaussian functions.
    • Each Gaussian explains the data present in each of the clusters
    available. 
    • Since there are three( k=3) clusters and the probability density is
    defined as a linear function of densities of all these k distributions.
    • As there will be some n number of sample points in k(th) cluster
    and the parameters cannot be estimated in closed form.
    • Now question is that how will you find out the missing or hidden data
    points? 

    • So the answer is the Expectation-Maximization(EM) algorithm.


    Expectation-Maximization(EM) Algorithm
    • What Expectation-Maximization(EM) algorithm does and how is it
    helpful for us?
    • It finds the maximum-likelihood estimates for model parameters when the
    data is missing or incomplete or has some variables hidden.
    • This algorithm chooses some random values for the missing data points and
    calculates a new set of data. These values are used to filling the missing points
    until the values get first.
    • This is how we will get all the data points of the cluster and the clusters will
    be formed.
    • So we can form the clusters using the Gaussian Mixture Model.
    • EM alternates between performing an expectation E-step, which
    computes an expectation of the likelihood by including the latent
    variables as if they were observed, and a maximization M-step, which
    computes the maximum likelihood estimates of the parameters by
    maximizing the expected likelihood found on the E-step.
    • The parameters found on the M-step are then used to begin another
    E-step, and the process is repeated until convergence.
    Applications of EM Algorithm
    • The latent variable model has several real-life applications in Machine
    learning:
    • Used to calculate the Gaussian density of a function.
    • Helpful to fill in the missing data during a sample.
    • Used for finding the values of latent variables.
    • Used in image reconstruction in the field of Medicine and Structural
    Engineering.
    • It finds plenty of use in different domains such as Natural Language
    Processing (NLP), Computer Vision, etc.
    • Used for estimating the parameters of the Hidden Markov Model (HMM) and
    also for some other mixed models like Gaussian Mixture Models, etc.
    Advantages of EM algorithm

    • The basic two steps of the EM algorithm i.e, E-step and M-step are
    often pretty easy for many of the machine learning problems in terms
    of implementation.
    • The solution to the M-steps often exists in the closed-form.
    • It is always guaranteed that the value of likelihood will increase after
    each iteration.
    Disadvantages of EM algorithm

    • It has slow convergence.
    • It is sensitive to starting point, converges to the local optimum only.
    • It cannot discover K (likelihood keeps growing with number of
    clusters)
    • It takes both forward and backward probabilities into account. This
    thing is in contrast to that of numerical optimization which considers
    only forward probabilities.
    References

    You might also like