초고속영상의 전배경 분리와 데이터 증강을 통한 말 자세 추정
Horse Pose Estimation using Background Subtraction and Data Augmentation of Ultra-speed Video
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 논문은 초고속영상을 통하여 말 자세를 추정하는 방법에 대한 연구이다. 미디어와 게임 산업의 발전으로 대상의 실제 움직임을 가상의 공간에서 표출하려는 연구가 급속히 증가하고 있다. 하지만 말은 사람에 비해 통제하기가 상대적으로 어려워 데이터 셋을 구축하는 데 제약이 많다. 이로 인해 데이터 셋이 매우 부족하여 연구의 발전 속도가 느리다. 이러한 문제를 해결하기 위해, 본 논문은 새로운 데이터 증강 방법을 제안하고자 한다. 먼저 고속으로 달리는 말을 초고속 카메라로 촬영하여 데이터 셋을 수집하고 전배경 분리와 영상 합성한다. 실험결과 MAE 0.2%~0.8%만큼 정확도가 개선되었고, 기존 Horse-10 데이터셋과 비교할 때 약 MAE 40%만큼 향상된 것을 확인했다.
Abstract
This paper is a study of horse posture estimation method using Ultra-high speed images. According to the development of media and game industries, researches to express the actual movement of an object in a virtual space are rapidly increasing. But horses are relatively difficult to control compared to people. There are many limitations in building data sets, so the advance of research is slow. To overcome this problem, this paper seeks to propose a novel data augmentation method. First, high-speed horses are photographed with ultra-high-speed cameras to collect data sets, and than background separation and image synthesis are performed. As the result of experiments, it was confirmed that the accuracy was improved by 0.2% to 0.8% of the MAE, and that it was improved by about 40% of the MAE compared to the existing Horse-10 dataset.
참고문헌 (15)
- Li, S., "ATRW:A benchmark for amur tiger re-identification in the wild", arXiv preprint arXiv:1906.05586, ~
- Mathis, A., "DeepLabCut : markerless pose estimation of user-defined body parts with deep learning", Nature neuroscience, 21(9), pp.1281~1289
- Nath, T., "Using DeepLabCut for 3D mar kerless pose estimation across species and behavio rs", Nature protocols, 14(7), pp.2152~2176 (2019)
- Krizhevsky, Alex, "Imagenet classification with deep convolution al neural networks", Advances in neural information processing systems, 25 (2012)
- He, K., "Deep residual learning for image recognition", Proceedings of the IEEE conference on computer vision and patter n recognition, pp.770~778 (2016)
- Howard, A. G., "Mobilenets: Effici ent convolutional neural networks for mobile vision applications", arXiv preprint arXiv:1704.04861, ~
- Tan, Mingxing, "Efficientnet : Rethinki ng model scaling for convolutional neural network s", International conference on machine learning, PMLR, pp.6105~6114 (2019)
- Li, S., "ATRW:A benchmark for amur tiger re-identification in the wild", arXiv preprint arXiv:1906.05586, ~
- Joska, D., "Acinoset : A 3d pose estimation dataset and baseline models for cheetahs in the wild", 2021 IEEE International Con ference on Robotics and Automation, pp.13901~13908
- Shooter, M., "SyDog : A Synthetic Dog Dataset for Improved 2D Pose Esti mation", arXiv preprint arXiv : 2108. 00249, ~ (2021)
- Zuffi, S., "Three-D Safari : Learning to Estimate Zebra Pose, Shape, and Texture from Images In the Wild", Proceedings of the IEEE, 2019, 5359~5368
- Goodfellow, I., "Generati ve adversarial nets", Advances in neural information processing systems, 27 (2014)
- Chen, L. C., "Encoder-decoder with atrous separabl e convolution for semantic image segmentation", Proceedings of the European conference on com puter vision (ECCV), pp.801~818 (2018)
- Sengupta, S., "Background mattin g : The world is your green screen", Proceedings of the IEEE, 2020, 2291~2300
- Frucci, M., "International Conference on Image Analysis and ProcessingSpringer", In International Conference on Image Analysis and ProcessingSpringer, Berlin, Heidelberg, pp, pp.168~177 (2011)