Traffic Sign using
CNN
Midterm Report
Tô Văn Hiệu
Nguyễn Huy
Hoàng
Trần Đình Vinh
Group 2 -
Table of contents
01
04
02
05
03
06
Timeline Background Methodology
Results Discussion Conclusion
Timeline
01
• After two weeks, we have moved into phase 4 and done some work related to
model running and training.
• This week, we will get to the next phase which involves refining and optimizing the
model.
Cascade
Localization
02
Haar Cascade
Haar Cascade is an algorithm created based on those features to detect objects (can be faces,
eyes, hands, objects, etc.) published in 2001 by Paul Viola and Michael Jones in the article Their
report claims “Fast object detection using cascades to enhance simple features.”.
Haar Cascade uses Haar layers and then uses many of those features over many passes
(Cascade) and creates a complete facial recognition machine.
Haar Features
To find out the regions containing traffic signs we used a well-known machine learning technique called
Haar Cascade Classifier. The features learned are contained in the output cascade.xml which is used by
OpenCV to find out the Region of Interests (ROI) that might contain traffic sign.
Localization
Introduction
Steps to create file cascade
(XML)
Apply Cascade XML
Using the Haar Cascade Classifier to localize objects like traffic signs has several good
reasons:
• Efficiency and Speed
• High accuracy
• Ease of training
• Multi-level detection capability
Model
structure
03
Model structure
Model
evaluation
04
The difference between goodfit, underfit and overfit model

Traffic Sign recogintion using CNN-ComputerVision_Group3_Midterm.pptx

  • 1.
    Traffic Sign using CNN MidtermReport Tô Văn Hiệu Nguyễn Huy Hoàng Trần Đình Vinh Group 2 -
  • 2.
    Table of contents 01 04 02 05 03 06 TimelineBackground Methodology Results Discussion Conclusion
  • 3.
  • 4.
    • After twoweeks, we have moved into phase 4 and done some work related to model running and training. • This week, we will get to the next phase which involves refining and optimizing the model.
  • 5.
  • 6.
    Haar Cascade Haar Cascadeis an algorithm created based on those features to detect objects (can be faces, eyes, hands, objects, etc.) published in 2001 by Paul Viola and Michael Jones in the article Their report claims “Fast object detection using cascades to enhance simple features.”. Haar Cascade uses Haar layers and then uses many of those features over many passes (Cascade) and creates a complete facial recognition machine. Haar Features
  • 7.
    To find outthe regions containing traffic signs we used a well-known machine learning technique called Haar Cascade Classifier. The features learned are contained in the output cascade.xml which is used by OpenCV to find out the Region of Interests (ROI) that might contain traffic sign. Localization Introduction
  • 8.
    Steps to createfile cascade (XML)
  • 9.
  • 10.
    Using the HaarCascade Classifier to localize objects like traffic signs has several good reasons: • Efficiency and Speed • High accuracy • Ease of training • Multi-level detection capability
  • 11.
  • 12.
  • 13.
  • 14.
    The difference betweengoodfit, underfit and overfit model