Adaptive auxiliary input extraction based on vanishing point detection for distant object detection in high-resolution railway scene

Abstract

Object detection plays an important role in intrusion detection of railway safety monitoring system. Generally, high-resolution image has to be down-sampled in order to reduce the amount of computation, resulting in missed detections of distant objects. This paper proposes an auxiliary input framework based on vanishing point (VP) detection to preserve the image resolution of distant objects in railway alarm region. We improved the VP detection network based on CNN classification, which consists of two branches, estimating the x-y coordinates respectively. Auxiliary input based on VP can improve the accuracy of target detection, especially for distant targets. Experiments on public data sets show that the proposed model overperforms single-branch model and regression model.