A pedestrian detection method using the extension of the HOG feature

Abstract

Development of an ITS (Intelligent Transport System) has drawn much attention from computer vision community in recent years. In particular, various techniques for detecting pedestrians automatically have been proposed by many researchers. Among them, the HOG feature proposed by Dalal & Triggs has gained much interest in the pedestrian detection. However, previous methods including the original HOG feature have not achieved satisfactory detection rates.In this paper, we propose an extension of the HOG feature, i.e., flexible choice of the number of bins and automatic definition of a cell size and a block size by parameterizing their scales. By comparative experiments, it was confirmed that the proposed method outperforms the previous methods in the performance of pedestrian detection.

Development and Characterization of a Computer Vision System for Human Body Detection and Tracking under Low-light Condition

“…The pedestrian dataset resulting from their works were made available to the public where algorithms on pedestrian detection can be trained and validated. [5], [6] and [7] made optimization on the algorithm to reduce computation time and improve accuracy. In [8], an in-depth analysis of the HOG -SVM chain answering the question why HOG-SVM performs so well in people detection was made.…”