camouflage errors on moving people D. Conte, P. Foggia, G. - - PowerPoint PPT Presentation
An algorithm for recovering camouflage errors on moving people D. Conte, P. Foggia, G. Percannella, F. Tufano, and M. Vento University of Salerno Italy Outline Definition of the problem Related works The proposed algorithm
30 August, 2010 An algorithm for recovering camouflage errors on moving people 2
Definition of the problem Related works The proposed algorithm Experimental results
30 August, 2010 An algorithm for recovering camouflage errors on moving people 3
Current Frame Background Foreground Mask Moving Objects
30 August, 2010 An algorithm for recovering camouflage errors on moving people 4
Background subtraction technique is subject to a set of
One of this is the camouflage
30 August, 2010 An algorithm for recovering camouflage errors on moving people 5
The effect of camouflage consists in a (random)
30 August, 2010 An algorithm for recovering camouflage errors on moving people 6
Camouflage has received less attention than the other
Some methods (e.g. [4]) involve the use of depth information
High computation cost and impracticable with legacy cameras
Other methods (e.g. [3, 8, 10]) propose some techniques to
A single camouflaged pixel is unrecognizable without analyze its
neighborhood
30 August, 2010 An algorithm for recovering camouflage errors on moving people 7
1.
1.
30 August, 2010 An algorithm for recovering camouflage errors on moving people 8
We focused our attention to the
The model must be carefully
A too detailed model would result
in many missed detection
A too general model would cause
the generation
many false positive errors
b1 and b2 are minimum and maximum real width h1 and h2 are minimum and maximum real height
30 August, 2010 An algorithm for recovering camouflage errors on moving people 9
Objective
grouping two or more blobs in order to form a unique blob
The procedure
repeatedly merging couples of blobs into larger ones until the
30 August, 2010 An algorithm for recovering camouflage errors on moving people 10
Conditions to group two blobs (represented by their bounding boxes X and Y)
1.
The projection on the horizontal axis of the bounding boxes are
2.
The real height of the grouped box is included between ℎ1 and ℎ2
3.
The real height of the grouped box is included between ℎ1 and ℎ2
p p p p
2 1 h
2 1 b
The grouped box Z is built starting from X and Y pixels coordinates Inverse Perspective Mapping is applied in order to determine its real size
30 August, 2010 An algorithm for recovering camouflage errors on moving people 11
Source image and a detail of a person Foreground Mask Bounding Boxes
30 August, 2010 An algorithm for recovering camouflage errors on moving people 12
The proposed method
is independent from the foreground detection algorithm used is used as a postprocessing on the output of four well known
Mixture of Gaussians [5] (MOG) Enhanced Background Subtraction [2] (EBS) Self-Organizing Background Subtraction [7] (SOBS) Statistical Background Algorithm [6] (SBA)
30 August, 2010 An algorithm for recovering camouflage errors on moving people 13
Performance index
Fuzzy definition for True Positive, False Positive and False Negative G set of bounding boxes of the ground truth where D set of bounding boxes detected by the algorithm | · | area of a bounding box
30 August, 2010 An algorithm for recovering camouflage errors on moving people 14
Dataset
videos acquired in different lighting and weather conditions NA1 NA3 NA2 Public database (PETS Conference)
http://www.cvg.rdg.ac.uk/PETS2006/
PETS MSA Indoor sequence (presented in [7])
30 August, 2010 An algorithm for recovering camouflage errors on moving people 15
30 August, 2010 An algorithm for recovering camouflage errors on moving people 16
We
Experimentations confirmed the effectiveness of the
Future improvements
Refining the model Definitions of suitable models for other objects of interest