I am using the HOGDescriptor class in OpenCV for object detection. It seem开发者_StackOverflows to me that the implementation only uses unsigned gradients, so it's unable to make the distinction between light->dark and dark->light transitions, a feature I would really need. Does anyone know an implementation of HOG that uses signed gradients?
Thanks!
I guess you can use the 'unsigned' and 'signed' parameter with OpenCV implementation, you only need specify the degrees. The orientation bins are evenly spaced over 0◦ – 180◦ (“unsigned” gradient) or 0◦ –360◦ (“signed” gradient).
精彩评论