Efficient Stop Sign Detection Using 3D Scene Geometry

1
Efficient Stop Sign Detection Using 3D Scene
Geometry

• Jeffrey Schlosser
• CS 223b
• March 17, 2008

2
Problem Statement and Data

• Quickly and reliably detect stop signs using data
  acquired from a camera mounted on an autonomous
  vehicle

3
Approach

• Haar classifiers used to form representation of
  stop signs
• Utilize known geometry

4
Approach (cont.)
Possible locations and scales of stop signs
5
Results Single Image
Traditional Algorithm (comprehensive search)
Intelligent Algorithm (using 3D scene context)
Processing time 1117 milliseconds
Processing time 275 milliseconds
6
Results Image Sequence
Intelligent Algorithm (using 3D scene context)
Traditional Algorithm (comprehensive search)

• Average Processing time 253 milliseconds
• 3.9 fps
• 1.53 FP/image

• Average Processing time 71 milliseconds
• 14 fps
• 0.73 FP/image

• Even faster times possible with better information

7
Questions?