Pedestrian detection

Abstract

In this work we study the problem of pedestrian detection in images and videos. At first we have analized the state of the art regarding the existing solutions, their implementations, and their strenghts and weaknesses. As result of this analysis we then focus on the approch based on Histogram of Oriented Gradients (HOG), proposed by Dalal and Triggs, which significantly outperforms existing algorithm for human detection, ensuring invariance to illumination, shadows and rotations of the images.

The key point of this approch is the costruction of the gradients histograms, which lead to obtain a great descriptor for the recognition task of pedestrian. At each step we also paid attention at the computational costs of the adopted solutions, trying to provide a method that combines precision and speed.

In order to evaluate performances we use dierent pedestrian databases under varying evironmental variables as: brightness, contrast, obstructed parts, variations in position and inclination, with fair results.