An Enhanced Histogram of Oriented Gradients for Pedestrian Detection

The outstanding Histogram-of-Oriented-Gradients (HOG) feature proposed by Dalal and Triggs is a state-of-art technique for pedestrian detection, and it is usually applied with a linear support vector machine (SVM) in a sliding-window framework. Most other algorithms for pedestrian detection use HOG as the basic feature, and combine other features with HOG to form the feature set. Hence, the HOG feature is actually the most efficient and fundamental feature for pedestrian detection. However, the HOG feature cannot adequately handle scale variation of pedestrians. In addition, simply downsampling an image into a different scale, or decomposing via wavelet into multi-resolution subimages, calculating their HOG feature and combining them cannot enhance performance. Therefore, in this paper, based on the idea of multi-resolution feature descriptors, we propose a new robust edge feature referred to as Enhanced HOG (eHOG). It is a complementary descriptor for the histograms-of-oriented-gradients feature. Though the extraction process of the eHOG descriptor is derived only from HOG itself, similar to the process of extracting edge information from the downscaling image, it retains much more information for the edge gradient than that of the original HOG, without significantly increasing the complexity of computation. In the INRIA pedestrian dataset, many experiments have been conducted with eHOG and HOG, and the results show that the proposed new feature consistently improves the detection rate more than the original HOG feature detector. Particularly, eHOG with a Histogram Intersection Kernel SVM (HIKSVM) classifier has greatly improved performance. These results suggest that eHOG may be a better substitute for HOG for pedestrian detection in many applications.