CLASSIFICATION OF INSPECTION ACTION OF ENGINEERS FOR EFFICIENT VIDEO PRESENTATION TO ASSIST ASSESSMENT OF INFRASTRUCTURE FACILITY DISTRESS

Abstract

In this study, we propose a method for classifying inspection actions of engineers to efficiently search for necessary scenes from inspection videos for the evaluation of distresses in infrastructure facilities. The inspection actions of engineers are more specific than basic actions such as walking and standing, and it is difficult to prepare a large dataset in advance. Therefore, this study proposes a novel classifier for inspection actions using the classification results of basic actions output from a action classification model that has already been trained on a large dataset. Furthermore, by using acton features, object features, and acoustic features, it is feasible to classify inspection actions focusing on the tools used for inspection and the sounds. We demonstrate the effectiveness of the proposed method by quantitatively verifying the accuracy of the inspection action classification and by presenting the classification results to verify the practicality.