Context-aware Pedestrian Detection with Salient Region Self-growing in Far-infrared Images

Abstract

Abstract-In this paper, we present a new framework to detect pedestrians in infrared images. The framework consists of a candidate generation module and a classification module, both of which are implemented based on convolution neural network. Specifically, we learned effective segmentation threshold by deep learning methods, and proposed a salient region self-growing algorithm to generate candidates. Besides, we conducted contextaware classification on the candidates to reduce the false positives using cues from the context. We achieved state-of-theart result on a public dataset, which has shown the effectiveness of the proposed method.