Vehicle Detection with Satellite Images

1
Vehicle Detection withSatellite Images

• Presented by
• Prem K. Goel
• NCRST-F, The Ohio State University
• Workshop on
• Satellite Based Traffic Measurement
• Berlin, Germany
• 9-10 September 2002

2
Image Processing Algorithms Performance
Evaluation

• Acknowledgment
• C. Merry, G. Sharma, F. Lu,
• M. McCord,
• Past students P. Goel, and J. Gardar

3
Vehicle Identification in High Resolution
Satellite Imagery

• Infrequent Image Acquisition from satellites
• Stereo Coverage May be Unavailable

4
IKONOS Satellite Imagery Tucson, AZ
5
Zooming-in
6
(No Transcript)
7
(No Transcript)
8
Image Segment for Processing
9
Zoomed and Pan Satellite Imagery (Columbus)
10
Problem Statement

• 1-m resolution image
• 8 or 11-bit data
• To detect and count vehicles
• Vehicle classes cars and trucks
• No road detection

11
Pavement Background Image

• Lack of stereo Images
• Background (Pavement) Image
• No Background

• Background Based
• Bayesian Background Transformation (BBT)
• Principal Components (PCA)
• Gradient Based

12
BBT Method Flow Chart
Highway Image (I)
Background (B)

Distributions of gray-levels in two
classes Initial prior probabilities

• Estimate probability of a pixel being stationary
  based on change from background

Background Transform
Estimate Distribution Parameters
Update probabilities
No
Converged?
Yes
Threshold
Clustering and other operations
Vehicle Counts
13
Principal Components (PCA) Method

• PCA-based Method
• Bands to capture texture and change
• Re-orient bands

14
Gradient based method

• Gradient Based Method
• The edge at vehicle boundaries
• Gradient image image with two classes

• Threshold
• try to incorporate spatial distribution of gray
  values

15

Final Outcome
Original Image
Binary Image
16
Simulated Images

• No Method was best
• Different method performed well for different
  images
• Performance Evaluation on Real Images crucial

17
Real Image Test Cases

• General Characteristics
• Vehicles vs. pavement
• pavement type, vehicle color, atmospheric
  conditions
• Objects Road signs, Lane markings
• Road geometry
• Traffic density

18
Image I 75 1

• Main Characteristic
• Pavement material transition

19
I 75 1
Probability Map
Clustered Probability Map
20
Image I 75 2

• Pavement material transition

21
I 75 2
Clustered Probability Map
Probability Map
22
Image I 270 1

• Pavement material transition
• Overpass
• Lane markings
• Curved road segment

23
I 270 1
Clustered Probability Map
Probability Map
24
Image I 270 2

• Lane markings
• Pavement material transition
• Straight segment
• Fairly dense traffic

25
I 270 2
Probability Map
Clustered Probability Map
26
Image I 70 1

• Lane markings
• Sign board
• Fairly dense traffic
• Straight road segment

27
I 70 1
Probability Map
Clustered Probability Map
28
Image I 10 1

• Straight road segment
• Median
• Good vehicle vs. pavement contrast

29
I 10 1
Clustered
Probability Map
30
Image I 270 3

• Multiple pavement material transitions
• Median
• High traffic density

31
I 270 3
32
Image I 71 1

• Poor vehicle vs. pavement contrast
• Illumination change
• Overpass

33
I 71 1
Clustered
Thresholded Gradient Img
Clustered
Probability Map
34
I 71 1
35
Image I 70 2

• Cloud cover
• Overpass
• Pavement material transition

36
I 70 2
Thresholded PC Band
Clustered
37
I 70 2
Thresholded Gradient Img
Clustered
38
I 70 2
Clustered Probability Map
Probability Map
39
I 70 2
40
Results Summary
Summary Errors of Omission and Commission

• BBT and gradient method give numbers close to the
  real values
• Large errors of omission and commission for PCA
  and gradient based method
• Low omission and commission errors for BBT method

41
Summary
42
Future Needs

• Methods Not Requiring Background
• Post-processing
• sieving and clustering
• Effort
• Process