Solichah, Uti (2020) Markerless Motion Capture Berbasis Openpose Model Menggunakan Metode Triangulasi. Masters thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
07111750052002-Master_Thesis.pdf - Accepted Version Download (18MB) | Preview |
Abstract
Human Pose Estimation merupakan salah satu topik menarik pada visi komputer. Dalam penelitian sebelumnya, Openpose telah berhasil menerapkan human pose estimation dalam bentuk 2D. 3D pose estimation didapatkan dari dua macam masukan, yaitu single view dan multiview, 3D pose estimation menggunakan multiview dapat mengestimasi lokasi kedalaman lebih robust dibandingkan dengan menggunakan masukan single view. 3D human pose estimation dapat dihasilkan dari dataset pose 3D atau dari lokasi joint 2D. Penggunaan dataset pose 3D memiliki keterbatasan dalam sumber maupun kegunaannya. Dari beberapa kasus di atas, Kami memilih untuk mengembangkan pendekatan multiview dan lokasi joint 2D sebagai masukan untuk mendapatkan 3D human pose estimation.
Input dari penelitian berupa dua view image yang berbeda, dimana masing-masing input akan diproses menggunakan inference openpose model untuk mendapatkan lokasi joint 2D. Proses kalibrasi kamera dibutuhkan untuk mendapatkan fitur instrinsik dan ekstrinsik kamera. Dengan fitur tersebut dan hasil dari 2D pose estimation, kita akan mendapatkan 3D pose estimation dengan pendekatan metode triangulasi. Sistem ini dapat digunakan dalam gender, baju dan pose yang berbeda. Desain jarak antar kamera yang terbaik pada jarak 66cm. Rentang error jarak dari kamera ke objek berkisar pada 13-21 cm. Error kedalaman objek ke kamera dapat diminimalkan dengan perbaikan hasil lokasi 2D joint dari openpose.
================================================================================================================================
Human pose estimation is one of many topics in computer vision. Previous works in openpose are only applicable in 2D human pose estimation. 3D human pose estimation can be done using two types of input, single and multiview. 3D pose estimation using multiview are inherently more robust than single view, due to multiview allowing better depth estimation. 3D human pose estimation are obtained from 3D data set poses or 2D joint location poses. However, with the limitations inherent in the 3D data set, such as lack of sufficient data and usage difficulties, this research applies 2D joint location. From the cases outlined before, we choose to develop multiview camera and 2D joint location as input to obtain 3D motion capture.
The inputs are two images with different views, where each image is processed using an inference openpose model to get 2D joint location. Camera calibration is needed to precisely obtain the intrinsic and extrinsic features of the camera. With these features and the 2D joint location results, we obtained the 3D human pose estimation using triangulation method. This system can be carried out on any combination of genders, apparels and poses. The best distance between cameras is 66 cm. The error distance between the camera and object range is 13-21 cm. Error in object distance to the camera can be minimalized by fixing 2D joint location from openpose.
| Item Type: | Thesis (Masters) |
|---|---|
| Additional Information: | RTE 621.367 Sol m-1 |
| Uncontrolled Keywords: | 2D human pose stimation, 3D human pose estimastion, kalibrasi kamera, multiview, openpose |
| Subjects: | Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.9 Computer algorithms. Virtual Reality. Computer simulation. T Technology > T Technology (General) > T57.6 Operations research--Mathematics. Goal programming T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis |
| Depositing User: | Uti Solichah |
| Date Deposited: | 08 May 2023 04:55 |
| Last Modified: | 08 May 2023 04:55 |
| URI: | http://repository.its.ac.id/id/eprint/73397 |
Actions (login required)
 |
View Item |