CS676: People counting GroupE

1
 CS676 People counting Group-E

• Guide Dr Amitabha Mukerjee
• Sourabh Daptardar (Y7111009)
• Makarand Gawade (Y7111022)

15th November 2007
2
Problem statement

• People Counting
• Identify number of people in a frame.
• Use tracking to count people across a sequence of
  frames.

3
Motivation

• Human Intrusion detection.
• Tracking usage of Resource / Preferences of
  people. (Unobtrusive monitoring)
• Optimizing working of road crossing signals.
• Getting a rough count of the number of people in
  an enclosed area (malls bank).

4
Methodology

• Build Template Database
• Background subtraction
• Cropped Images are labeled
• Find Haar Transforms
• Matching
• Find Haar of Input Image
• Pyramidal matching with Templates
• Maintaining count of people

5
Template Database Creation Tool
Select. templates
Resize Templates.
Input /Output directory
Navigation
6
Build Template Database

• Background subtraction

Crop FG object using GMM.
7
Build Template Database

• Cropped Images are labeled

Label Templates
8
Build Template Database

• Find Haar Transforms

-1 1
Horizontal Haar
-1 1
Vertical Haar
-1 1 1 -1
Diagonal Haar
9
Matching
Input Image
10
Matching
Perform HAAR transform on input image.
11
Matching
Find a match for the template in the image,
using Pyramidal search
12
Matching
Multi resolution decomposition allows templates
to be scaled while matching
13
Results
True ve 6
False ve 0
Actual 11
Match
Accuracy 55
Figure 1 LHCVideo1_002000.jpg
14
Results
Occlusion
True ve 5
False ve 0
Actual 11
Accuracy 45
Figure 2 LHCVideo1_002001.jpg
15
Results
True ve 4
False ve 0
False negative
Actual 11
Accuracy 36
Figure 3 LHCVideo1_002002.jpg
16
Results
True ve 5
False ve 0
Actual 13
Accuracy 39
Figure 4 LHCVideo1_005000.jpg
17
Results
18
Analysis

• The method used for detecting People gave a
  variable accuracy across image frames.
• As we move away from the frame from which
  templates were extracted, we see increased False
  negatives.
• The accuracy of the detection was found to be
  around 31 percentage.
• However it needs to be noted that there were only
  a very small percentage of False positives.
  Indicating that the threshold could be made less
  stringent to decrease False negatives.
• Also the number of templates used was very less,
  just 85, with more templates we could have seen
  better results.

19
Experiments

• Used two different Haar features.
• Work on Gray scale image.
• Using Pearson's coefficient for template
  matching.

20
Issues

• Occlusion
• People far away from the camera are two small to
  be detected reliably. Reducing threshold/adding
  templates to detect them increased false
  positives.

21
Future directions

• Need to build a larger database of templates
• Efficient matching algorithm is required
• In a video frame sequence
• The number of people do not change drastically in
  adjacent frames
• Search space could be reduced
• Tracking people across frames

22
References

• Michael Oren, Pedestrian Detection Using Wavelet
  Templates in Computer Vision and Pattern
  Recognition, 1997. Proceedings., 1997 IEEE
  Computer Society Conference.
• Paul Viola and Michael Jones, Robust Real-time
  Object Detection in International Journal of
  Computer Vision.
• Eric J. Stollnitz Tony D. DeRosey David H.
  Salesiny, Wavelets for Computer Graphics A
  Primer in Technical Report 94-09-11, Department
  of Computer Science and Engineering, University
  of Washington