Pr

1
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
VEHICLES DETECTION FROM AERIAL SEQUENCES Center
of Robotics, Electrical engineering and Automatic
- EA3299 University of Picardie Jules Verne CREA,
7 rue du moulin neuf 80000 Amiens,
France Diagnosis and Advanced Vehicles (DIVA)
Pole Conseil Régional
de Picardie
University of Picardie Jules Verne


1
2
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Aerial sequences Analysis taken from an
UAV-Camera system Proposed approaches aim to
extract and recognize vehicles in the road
University of Picardie Jules Verne


2
3
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting

• Using computer vision tools gt A large basis
  of information
• A whole description of the traffic
• Vehicle counts
• Vehicle speed
• Vehicle density
• Flow rates etc.
• Road traffic monitoring
• Congestion and incident detection
• Law enforcement
• Automatic vehicle tracking etc.

University of Picardie Jules Verne


3
4
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting

• Computer vision systems for road traffic
  monitoring
• Static vision system fixed camera
• Dynamic vision system moving camera

University of Picardie Jules Verne


4
5
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Static vision system gt Fixed background
Approaches farm the difference between
acquired images and background Moving
vehicles are ,so, extracted
University of Picardie Jules Verne


5
6
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Dynamic vision system Camera-UAV
system Having a fixed background is
impossible
University of Picardie Jules Verne


6
7
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting

• We propose two approaches to extract vehicles
• Approch based on perceptual (geometrical)
  organization
• Approch based on  common fate  principle

University of Picardie Jules Verne


7
8
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on perceptual (geometrical)
organization
University of Picardie Jules Verne


8
9
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on perceptual (geometrical)
organization

• A graph problem where
• Nodes are images edges.
• Links based on two criteria
• Parallelism
• Proximity

University of Picardie Jules Verne


9
10
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on perceptual (geometrical)
organization
University of Picardie Jules Verne


10
11
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on  common fate  Principle

• Sequences taken from an UAV-camera system
• Two types of movement
• objects movement or displacement (in our case
  edges presenting vehicles)
• background movement.
• The idea distingush between these two kinds of
  movement

University of Picardie Jules Verne


11
12
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on  common fate  Principle
Corners Detection Image(t),Image(t1)
Primitives Detection Image(t)
Primitives Description
Matching
Displacements Computation
Homogeneous Primitives Extraction
rank(W) ?
Results Verifying ?
University of Picardie Jules Verne


12
13
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on  common fate  Principle

• Why do we use corners data to matching images
  edges ?
• Corners matching process is less complicated than
  edges matching process
• Corners rate repeatability is more elevated than
  edges rate repeatability
• Edge displacement is computed as the mean of
  corners displacements, so false matching effects
  are reduced

University of Picardie Jules Verne


13
14
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on  common fate  Principle
Matching tool Computing of the Mahalanobis
distances between corners invariants vectors.
University of Picardie Jules Verne


14
15
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on  common fate  Principle

• Homogeneous Primitives Extraction ?
• A graph problem where
• Nodes are images edges
• Links are nodes displacements similarity

University of Picardie Jules Verne


15
16
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on  common fate  Principle
Partitioning tool Normalized cuts
technique Link between two nodes (edges) i and j

University of Picardie Jules Verne


16
17
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on  common fate  Principle
Verifying Algorithm The Dempster Shafer
Theory Verifying system has 5 input sensors
and 3 output degrees
University of Picardie Jules Verne


17
18
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on  common fate  Principle
V 79,95 Conflict 1
University of Picardie Jules Verne


18
19
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on  common fate  Principle
NR Number of Rejected classifications before
converging
University of Picardie Jules Verne


19
20
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
Approch based on  common fate  Principle
NR Number of Rejected classifications before
converging
University of Picardie Jules Verne


20
21
Vehicles Detection From Aerial Sequences
4th European Micro-UAV Meeting
THANK YOU !
University of Picardie Jules Verne


21