Enhancement of fast pedestrian detection using HOG features

• Main Authors: Baghdadi, Ahmad, Abd Latif, Suhaimi
• Format: Article
• Language: English
• Published: Institute of Research and Journals 2015
• Subjects: T10.5 Communication of technical information
• Online Access: http://irep.iium.edu.my/44555/1/11-126-142960909718-22.pdf; http://irep.iium.edu.my/44555/; http://pep.ijieee.org.in/journal_pdf/11-126-142960909718-22.pdf
• Institution: Universiti Islam Antarabangsa Malaysia

Histograms of Oriented Gradients (HOG) suffers low processing speed, due to complex and redundant computations. In this paper, a method to improve the speed of detection while maintaining the accuracy has been proposed.This is implemented without using any special purpose hardware such as GPU. The speed has been substantially increased through the calculations of trilinear interpolation within each block of cells, utilizing the sub-cell method. The redundant calculation over two levels has been avoided: within each block and within the detection window by utilizing the concept of reusing the calculated features. Results showed a significant speed up by more than eleven times, with almost the same detection accuracy of the original algorithm.