딥러닝 기반 객체 검출기와 스펙트럼 군집화를 이용한 군용객체 몸체 추정
Detector and Spectral Clustering
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 연구는 딥러닝을 바탕으로 군용객체를 검출하고, 포신을 갖고 있는 객체에 대해 포신을 제외한 몸체 영역을 추정하는 시스템을 개발한다. 이를 위해 전차, 장갑차, 사람 객체의 클래스로 구성된 데이터셋을 구축한다. 특히, 전차 클래스는 주로 원격조종 전차를 촬영하여 얻은 데이터로 구축되고, 전체 데이터셋은 객체 탐지를 위해 YOLOX 검출기의 학습과 테스트에 이용된다. 전차 클래스로 검출된 객체는 검출된 영역을 분리하고 영역의 하단 부분을 여러 개의 셀로 나눈다. 그 후, 각 셀의 RGB 히스토그램의 분포를 계산하고 셀 간의 유사도를 통한 스펙트럼 군집화를 적용하여 몸체와 배경을 분류한다. YOLOX-L 모델을 사용하였을 때 IoU 0.5에서 89.2%의 AP 결과를 얻었으며, 포신을 제외한 군용객체의 몸체 부분을 인식하는 것을 보였다. 본 연구에서 개발된 기술을 바탕으로 방위 시나리오에서의 객체 탐지 능력을 향상시키고, 상대 전차의 몸체 부분을 바탕으로 포탑 구동을 통해 정확한 조준 및 포격하는데 활용된다.
Abstract
This study develops a system that detects military objects based on deep learning and estimates the bodies of military objects excluding gun barrels. For this purpose, we build a dataset consisting of classes of tanks, panzers, and human objects. In particular, the tank class mainly consists of data obtained by capturing remotely controlled tanks, and the entire dataset is used for training and testing of YOLOX detector for object detection. An object detected as the tank class is segmented into a regions of interests, and the lower part of the image is divided into several cells. After the distribution of the RGB histogram in each cell is computed, spectral clustering through similarity between histograms of cells is applied to classify the body and background. For YOLOX-L, we achieve 89.2% , and show that the body parts of military objects excluding the gun barrel are well recognized. The proposed system not only improves the ability to detect objects in defense scenarios, but also accurate aiming and artillery fire are achieved by driving the turret based on the body parts for opposing tanks.
참고문헌 (23)
- Jinyong Hwang, "A Study on Transfer Learningbased Ensemble Model for Classification of Maneuver Weapon System", The Journal of Korean Institute of Information Technology, 19(12), pp.1~10 (2021)
- Wonjae Lee, "Study on Upgrading the Scientific Boundary Monitoring System : Focusing on Image Data Analysis", Journal of the Korea Academia-Industrial cooperation Society, 23(6), pp.144~150 (2022)
- Seunggyun Lim, "Identifications and Evaluation of Tank Nationality using YOLO Algorithm", KIISE Transactions on Computing Practices, 27(12), pp.555~562 (2021)
- M. Sandler, "Mobilenetv2 : Inverted residuals and linear bottlenecks", Proceedings of the IEEE conference on computer vision and pattern recognition, pp.4510~4520 (2018)
- K. He, "Deep residual learning for image recognition", Proceedings of the IEEE conference on computer vision and pattern recognition, pp.770~778 (2016)
- G. Huang, "Densely Connected Convolutional Networks", Proceeding of IEEE Conference on computer vision and pattern recognition, pp.4700~4708 (2017)
- C. Szegedy, "Rethinking the inception architecture for computer vision", Proceedings of the IEEE conference on computer vision and pattern recognition, pp.2818~2826 (2016)
- J. Deng, "Imagenet : A large-scale hierarchical image database", 2009 IEEE conference on computer vision and pattern recognition, pp.248~255 (2009), Ieee
- T. Y. Lin, "Microsoft coco : Common objects in context", Computer Vision–ECCV 2014:13th European Conference, 13, pp.740~755 (2014), Springer International Publishing
- Gyeong-Jae Lee, "A Study on Improving Weapon System Detection Performance Using Sim2Real Learning", Journal of the Korea Academia-Industrial cooperation Society, 24(7), pp.189~196 (2023)
- U. Von Luxburg, "A tutorial on spectral clustering", Statistics and computing, 17, pp.395~416 (2007)
- M. Everingham, "The pascal visual object classes(voc)challenge", International journal of computer vision, 88, pp.303~338 (2010)
- J. Lee, "Local texture estimator for implicit representation function", Proceedings of the IEEE, 2022, 1929~1938
- J. Redmon, "You only look once : Unified, realtime object detection", Proceedings of the IEEE conference on computer vision and pattern recognition, pp.779~788 (2016)
- W. Liu, "Ssd : Single shot multibox detector", Computer Vision–ECCV 2016: 14th European Conference, 14, pp.21~37 (2016), Springer International Publishing
- R. Girshick, "Rich feature hierarchies for accurate object detection and semantic segmentation", Proceedings of the IEEE conference on computer vision and pattern recognition, pp.580~587 (2014)
- R. Girshick, "Fast r-cnn", Proceedings of the IEEE international conference on computer vision, pp.1440~1448 (2015)
- S. Ren, "Faster r-cnn : Towards real-time object detection with region proposal networks", Advances in neural information processing systems, 28 (2015)
- K. He, "Mask r-cnn", Proceedings of the IEEE international conference on computer vision, pp.2961~2969 (2017)
- H. Fan, "Lasot : A high-quality benchmark for largescale single object tracking", Proceedings of the IEEE, 2019, 5374~5383