Description: The system presents a new framework for traffic density estimation based on topic model, which is an unsupervised model. It uses a set of visual features without any individual vehicle detection and tracking need, and discovers the motion patterns automatically in traffic scenes by using topic model.

1. Dept. of Electronics & Telecommunication Engg.
S. S .G. M. College of Engineering, Shegaon
A Presentation on
Presented By:
Nidhi Shirbhayye
Under the Guidance:
Prof. A. N. Dolas

2. •Contents:
 Introduction.
 Literature Review.
 Background Theory.
 Methodology.
 Conclusion.
 References.

3. Introduction:
 Traffic density estimation plays an integral role in
intelligent transportation systems (ITS) for controlling
signals.
 The system presents a new framework for traffic
density estimation based on topic model, which is an
unsupervised model.
 It uses a set of visual features without any individual
vehicle detection and tracking need, and discovers the
motion patterns automatically in traffic scenes by
using topic model.

4. Cont…
 Topic Model.
 Latent Dirichilet Allocation.
 Latent Motion Patterns.
 Optical Flow.
 Log-Likelihood.

5. Topic model:
Aim of Topic Models:
• Large unstructured
collection of document
• Discover set of topics
that generated the
documents
• Annotate documents
with topics

6. What is a Topic?
6
Topic: A broad concept/theme,
semantically coherent, which is
hidden in documents
Representation: a probabilistic
distribution over words.
retrieval 0.2
information 0.15
model 0.08
query 0.07
language 0.06
feedback 0.03
……
e.g., politics; sports; technology;
entertainment; education etc.
CS@UVa

7. General idea of probabilistic
topic models
 Topic: a multinomial distribution over words
 Document: a mixture of topics
 A document is “generated” by first sampling topics from
some prior distribution
 Each time, sample a word from a corresponding topic
 Many variations of how these topics are mixed
 Topic modeling
 Images can be represented in the vector space model, how
an underlying model can be learnt given a large number of
images and how this model can be applied to do interesting
inferences.
7

8. Latent Dirichelet Allocation:
 Latent Dirichlet Allocation (LDA) is a generative
statistical model that allows sets of observations to be
explained by unobserved groups that explain why some
parts of the data are similar.
•M-Number of Documents
•N-Number of Words
•β – word Probability Matrix
•α- parameter of Dirichlrt
distribution
•Ѳ- Topic distribution

9. Latent Motion Patterns:
 In many surveillance scenarios, such as monitoring traffic
at intersections, crowded video scenes with various
motions may be involved.
 In these scenes, some typical activities, called motion
patterns, occur regularly and periodically.

10. Optical Flow:
 Motion estimation generally known as optical or optic
flow.
 Optical flow or optic flow is the pattern of apparent
motion of objects, edges and surface in a visual scene
caused by the relative motion between an observer (an eye
or a camera) and the scene.

11. Log-Likelyhood:
 In informal contexts, "likelihood" is often used as a
synonym for "probability." In statistics, a distinction is
made depending on the roles of outcomes vs. parameters.
 Probability is used before data are available to describe
possible future outcomes given a fixed value for the
parameter (or parameter vector).
 Likelihood is used after data are available to describe a
function of a parameter (or parameter vector) for a given
outcome.

12. Literature Review:
In the past, road engineers tried to estimate the
traffic flow or traffic density on a road by using
magnetic loop detectors or supersonic wave
detectors, some operators manually estimate the
traffic density and many other that was so boring
and inefficient

13. 1. Magnetic Loop Detector:
 APPROACH
 One or more loops of wire are embedded under
the road & connected to a control box.
When a vehicle passes over or rests on the loop,
inductance is reduced showing a vehicle .

14. 2. Smart Video Survellience System For Vehicle
Detection and Traffic Flow Control :
 APPROACH
 Traffic-flow measurement and automatic incident
detection using video cameras is another form of
vehicle detection.
 Video from cameras is fed into processors that analyse
the changing characteristics of the video image as
vehicles pass.

15. 3. A real-time computer vision system for
vehicle tracking and traffic surveillance :
 APPROACH
Instead of tracking entire vehicles, vehicle
features are tracked to make the system robust
to partial occlusion.
After the features exit the tracking region, they
are grouped into discrete vehicles using a
common motion constraint.
 The groups represent individual vehicle
trajectories which can be used to measure
traditional traffic parameters as well as new
metrics suitable for improved automated
surveillance

16. 4. Real-time Area Based Traffic Density
Estimation by Image Processing for Traffic Signal
Control System.
 This project describes a method of real time area based
traffic density estimation. Area occupied by the edges of
vehicles will be considered to estimate vehicle density.

17. Methodology:
 ROI Determination
 Video Representation
 Model Learning
 Density Estimation

18. ROI Determination:
The purpose of selecting ROI is to exclude the
unnecessary background information such as other road
lane.
Fig. Traffic scene
Fig. ROI

19. Video Representation:
To perform the feature extraction, we first temporally
divide the entire video into Nd non-overlapping short
clips. Then corner detector is employed to extract the
key points.
For each pair of consecutive frames, key points are
used to discover the optical flows.

20. Model Learning:
 Topic model is used to discover the set of latent motion
patterns from video .
 These learned motion patterns are employed to calculate
likelihood measure to estimate traffic density in traffic
videos

21. Traffic Density Estimation:
 The topic model is trained with specific density.
 The clips with the same density as the training dataset will
produce high log-likelihood.
 We use two thresholds to determine type of the traffic
density

22. Advantages:
 No need to individual vehicle detection and tracking.
 High accuracy
 Reduces time.
 Reliable.
 Traffic density computation is easy.
 It can be used when large amount of visual data is
available.

23. Conclusion:
 Automatically classify complex traffic videos and
determine their traffic density, based on LDA, which is
one of the most successful topic models
 The execution time for this approach is relatively low,
it can be used in real-time application.