Bayesian Model to Validate Complicated Video

7/31/2019 Bayesian Model to Validate Complicated Video

1/41

Hierachical Bayesian Model to Validate

Unsupervised activity perception in crowded and

complicated scenes


2/41

In unsupervised learning framework to model activities and

interactions in crowded and complicated scenes.

Under our framework hierarchical Bayesian models are used to

connect three elements in visual surveillance:

low-level visual features

simple atomic activities

Interactions


3/41

Atomic activities are modeled as distributions over low-

level visual features, and multi-agent interactions are

modeled as distributions over atomic activities. These

models are learnt in an unsupervised way.

Given a long video sequence, moving pixels are clustered

into different atomic and short video clips are

clustered into different interactions.


4/41

Maa-yoke Technology:

Maa-Yoke has a great deal of experience and knowledge and can

assist organizations in the following areas:

IT Service Management

IT Sales, Service & Support

Engineering


5/41

SIFY

P&O NEDLOYD - PUNE

SONY INDIA LTD

FUTURE SOFT


6/41

The Existing system fall in two categories namely:

The Object of interest is first detected and tracked

The Object is classified into categories.

Here the tracking is made by human so that error may occur

during the process.

Then video clips are made into image frames and the decided

object is tracked. When error done in few frames the result may

be concluded wrong.


7/41

Next used approach is motion feature vectors to describe video

clip. The video clip is partition into bipartite graph.Here we dont use any detection or tracking, so an individual

activity cannot be separated so this lead to failure.


8/41

In the current system we uses Hierarchical Bayesian Model which

has three models:

Latent Dirichlet Allocation(LDA)

Hierarchical Dirichlet Process (HDP) Mixture Model

Dual Hierarchical Dirichlet process (Dual - HDP) Model.


9/41

Latent Dirichlet Allocation(LDA):

The LDA mixture model assumes that it is already known how

many different types of atomic activities and interactions occur in

the scene.

Hierarchical Dirichlet Process (HDP) Mixture Model:

The HDP mixture model automatically decides the number of

categories of atomic activities.

Dual Hierarchical Dirichlet process (Dual - HDP) Model.

The Dual-HDP automatically decides the numbers of categories

of both atomic activities and interactions.


10/41

Hardware Requirements

Processor : Pentium-IV

Speed : 1.1GHz

RAM : 512MB

Hard Disk : 40GBGeneral : Key Board, Monitor, Mouse

Software Requirements

Operating System : Windows XPSoftware : JAVA (JDK 1.5.0)

Protocol : UDP


11/41

Module Description:

There are four modules in our projects

Video Converted Into Frames

Find the Movable Object

Separation of Object

Apply the Rules in the Object


12/41

Automatically segment a continuous video sequence.

Video Segment consists of number of pixel image frames

Slice the particular segment with a fixed rows and columns

segment the frames.

Identify the human objects


13/41

To fix the pedestrian area or path to cross the movable

objects is called point safety place.

To compare each pixel frames and print the current

status of the pixel each moving pixel is labeled by its

location and direction of the objects.


14/41

First step we can select the horizontal or vertical directory

paste it and load it. then separated it, and split it.A long

video sequence we can segment it based on different types

of interactions.

Our models provide a natural way to complete this task in anunsupervised since videos clips are automatically separated

into clusters.


15/41

Apply the rules and find out the wrong user show the difference

between the vehicle and the human being based on the size and

speed.

we can get pedestrian area change movement and need to check if

the detected movements come in pedestrian area the human being

going in right path, or otherwise people going in wrong path.


16/41


17/41


18/41


19/41


20/41


21/41


22/41

Segmentation Page


23/41


24/41


25/41


26/41


27/41


28/41


29/41


30/41


31/41


32/41


33/41


34/41


35/41


36/41

To detect the abnormal activity in crowded place by

using Hierarchical Bayesian models

To overcome the systemize process in the crowded

places

To avoid the human error in crowded place activities

Example: Traffic signal is the best example of the project


37/41

In this framework, we adopt the positions and moving

directions of moving pixels as low-level visual features since

they are more reliable in a crowded scene.

While we have demonstrated the effectiveness of this model in

a variety of visual surveillance tasks, including more

complicated features is expected to further boost the models

discrimination power.


38/41

For example, if a pedestrian is walking along the path of

vehicles, just based on positions and moving detections, his

motions cannot be distinguished from those of vehicles and this

activity will not be detected as an abnormality.

If a car drives extremely fast, it will not be detected as

abnormal either. Other features, such as appearance and speed,

are useful in these scenarios.


39/41

We have proposed an unsupervised framework adopting

hierarchical Bayesian models to model activities and

interactions in crowded and complicated scenes.

Three hierarchical Bayesian models: the LDA mixture model,

the HDP mixture model, and the Dual-HDP model are

proposed.


40/41

Without tracking and human labeling, our system is able to

summarize typical activities and interactions in the scene,

segment the video sequences, detect typical and abnormal

activities, and support high-level semantic queries on activitiesand interactions. These surveillance tasks are formulated in an

integral probabilistic way.


41/41

THANKYOU

Bayesian Model to Validate Complicated Video

Documents