Pedestrian Detection using HOG Features

Goal: Implement pedestrian detection using HOG and libsvm. I am able to extract HOG features for each image, and train/test libsvm classifier successfully (I manually verified predictions that pedestrians were/were not present).

I need guidance on the following... In the image, I want to detect exactly where the pedestrian is, and create a bounding box around the pedestrian.

My guess is that I do a sliding window on the image, extract the HOG features from the window, test those features within libsvm, and draw a rectangle around the window if it comes back positive.

Am I on the right track? Any code hints or snippets would be helpful.

[Algorithms Analyst]

HI Schrute

Hello

can you tell me plz how did you apply the HOG function on image?

Answer 1

[Image Analyst]

See sections 16, 18, and 22 for articles by those who have done it before: http://iris.usc.edu/Vision-Notes/bibliography/contents.html

[Schrute]

This sparked some other thoughts, which eventually helped me figure out the answer/implementation. Thank you.