Efficient Human Detection Based on Parallelimplementation of Gradient and Texture FeatureExtraction Methods

Pedestrian Detection is of interest in many computervision applications such as intelligent transportation systems andhuman-robot interaction; among the existing methods, thecombination of shape feature (i.e. Histogram of OrientedGradients (HOG)) and texture features (i.e. Local Binary Pattern(LBP)) has shown promising results in detection accuracy, but itis limited due to computation cost. In this paper, we introduce anew pedestrian detection algorithm with fast computation ofthese features on GPU. We propose a robust and rapidpedestrian detector by combining the HOG with LBP, as thefeature set and corresponding Support Vector Machine (SVM)classifiers. Also, we use the integral image method and anefficient parallel implementation to reduce detection time. Wecan achieve a more than 10x speed up, and 7% increase indetection rate.

کلیدواژه ها:

Pedestrian Detection ، Histogram of OrientedGradient ، Local Binary Pattern ، Integral Image ، GraphicsProcessing Unit