THE CLASSIFICATION OF UNDERWATER OBJECT USING POINT CLOUD DEEP LEARNING WITH 3D CAD AND COMPUTER GRAPHICS LIGHTING

Abstract

 This paper presents an accurate and efficient method for object classification of underwater structures by applying a point cloud deep learning method to underwater point cloud data obtained from a narrow multibeam echo-sounder. Specifically, the paper presents an automatic method for generating point cloud training data from a 3D CAD model of an underwater structure by utilizing the CG lighting technology. Additionally, we added a cut model, which is a partial model of the CAD model, in order to improve the accuracy of object classification. The validity and effectiveness of the method are verified by applying the training data generated by the proposed method to PointNet++, a point cloud deep learning framework, and evaluating the classification accuracy of actual underwater objects.