chapter6

[VINOD BATTINI]

how to detect the signs using camera , in real time traffic detection , can u please provide me some hints how to make this? 

[Michael Beyeler]

Hi Vinod,

A good place to start would be to find a publicly available dataset that shows traffic signs in their natural surrounding. For example:

• Linkopings Universitet: http://www.cvl.isy.liu.se/research/datasets/traffic-signs-dataset
• LISA: http://cvrr.ucsd.edu/LISA/lisa-traffic-sign-dataset.html

You could then train a Haar cascade classifier on the images, by splitting the dataset into images that contain signs vs. other images that don't.

How to train Haar cascades is explained here: http://docs.opencv.org/3.2.0/dc/d88/tutorial_traincascade.html.

You would then hope that the trained Haar cascade can generalize to your dashboard camera in real-world traffic.

If you also wanted to classify the signs, you could crop out the detected traffic sign in every real-time frame, and run it through the Chapter 6 code.

If you are more committed to the project than that, you could collect your own dataset with the dashboard camera that will later be used in real-time traffic detection. Problem is you would have to annotate every frame by drawing a bounding box around all traffic signs. You could let others do the work by using Amazon Mechanical Turk.

Then train the Haar cascade on your own collected data. This way it's most likely that the resulting Haar cascade will generalize to the real-world setting.

Best,

Michael