An Implementation of Environment Recognition for Enhancement of Advanced Video Based Railway Inspection Car Detection Modules

Abstract

In this paper, we propose a method for determining video environmental conditions as a preliminary step toward a video-based monitoring system for Cu-alloy overhead contact line (OCL) components. Previous video-monitoring methods have often produced severe errors when applied to trains with long operations because of their frequently changing environments, presenting inconsistent image conditions. Therefore, we propose a method that utilizes a set of image features capable of describing the given environment of video data acquired from an actual train. This set of features is applied to an environmental classifier based on the support vector machine to select appropriate parameters for further processing. Then, the proposed method is evaluated in terms of detection performance and distance error. The experimental results confirm that robust performance is realized in practical conditions of railway operation and especially in the OCL-monitoring scenario.