Object detection with hough forests using HaarHOG descriptor
Seyyed Reza Tabasi, Mahdi Zarif · 2014
In this paper we propose a new method for detecting object class instances based on Hough transform. Hough forests which are adapted to perform Hough transform have been efficiently used for single-class object detection. In this work we extend them using HaarHOG descriptor which is a combination of Haar wavelet and HOG descriptor. As a result, we increase the number of feature channels in Hough forests. Our experiments demonstrate that the proposed method performs as well as the traditional Hough forests and can also improve the detection accuracy for certain values of detection parameter.