Improved Local Texture Features for Pedestrian Detection

Abstract

Pedestrian detection is a hot issue in the field of computer vision and image processing in recent years. It has important application value in the domain of unmanned cars and driver assistance systems and so on, but there are existed many problems that need to be solved. In this paper, we present an improved texture feature MLBP (Mean of Local Binary Pattern) and the CMLBP (Color based on Mean of Local Binary Pattern) feature based on various color spaces. When the uniform LBP feature does not consider the influence of noise, the mutation of central pixel and neighborhood pixel is not taken into account and therefore the extraction processes of MLBP feature improve the calculation method of the uniform LBP, which makes the extracted feature more stable. The MLBP feature is extracted from gray images, yet color images transformed into gray images generally loss a great amount of information. In view of this point, we also propose the CMLBP feature based on multiple color spaces that is a more comprehensive description of the texture feature of images. In the INRIA pedestrian dataset, many experiments have been conducted with SVM and HIKSVM classifier, and the results manifest that the detection rates of MLBP and CMLBP are much better than the uniform LBP and the basic LBP. The combination of MLBP, CMLBP and other features has been applied to pedestrian detection, which also achieves good results.