Pedestrian Detection Based on Improved Interface Difference and Double Threshold with Parallel Structure

Abstract

In this paper, based on the classical "two-frame difference method" and "three-frame difference method", an improved pedestrian detection scheme based on "three-frame difference method" is proposed. Two improvements were made based on the conventional three-frame difference method. The first is to improve the "adjacent frame difference method "of the original three-frame difference to "three-frame frame mutual difference", effectively avoiding the influence of pedestrian behavior on the system. The second is to introduce the "double threshold parallel structure" to transform the image, which relatively increases the useful information contained in the image, and increases the accuracy of the system.