講演抄録/キーワード |
講演名 |
2024-02-23 16:20
SVMを用いた角度と距離特徴に基づく人の行動認識に関する研究 ○Cho Nilar Phyo・Thi Thi Zin・Pyke Tin(Univ. of Miyazaki)・Hiromitsu Hama(Osaka City Univ.) HIP2023-112 |
抄録 |
(和) |
Human action recognition is the important research area in computer vision research area and popular due to its enormous advantages for the human society. The analysis of human behavior includes gesture recognition, human action recognition, human interaction recognition and the recognition of multiple or group people interaction. Many researches about the human recognition actions have come out during these days. But HAR still have a lot of challenges for solving to be able to apply in real-world. In this paper, we proposed simple and effective HAR using the angles and distance as main feature. In our proposed system, we use skeleton joint points information generated by Microsoft Kinect Version 2 sensor and then extract the angles and distance feature according to the joint point information. Finally, we apply SVM over those features in order to implement the human actions recognition system. We have performed a lot of experiments using the video data. According to our experiments, we found that our proposed method achieves the overall accuracy of 99.25% for recognizing the 7 actions and the computational cost of the proposed system is also suitable for implementation the real-time application. |
(英) |
Human action recognition is the important research area in computer vision research area and popular due to its enormous advantages for the human society. The analysis of human behavior includes gesture recognition, human action recognition, human interaction recognition and the recognition of multiple or group people interaction. Many researches about the human recognition actions have come out during these days. But HAR still have a lot of challenges for solving to be able to apply in real-world. In this paper, we proposed simple and effective HAR using the angles and distance as main feature. In our proposed system, we use skeleton joint points information generated by Microsoft Kinect Version 2 sensor and then extract the angles and distance feature according to the joint point information. Finally, we apply SVM over those features in order to implement the human actions recognition system. We have performed a lot of experiments using the video data. According to our experiments, we found that our proposed method achieves the overall accuracy of 99.25% for recognizing the 7 actions and the computational cost of the proposed system is also suitable for implementation the real-time application. |
キーワード |
(和) |
人の行動認識 / 深度カメラ / スケルトン関節点情報 / サポートベクトルマシン / / / / |
(英) |
人の行動認識 / 深度カメラ / スケルトン関節点情報 / サポートベクトルマシン / / / / |
文献情報 |
信学技報, vol. 123, no. 386, HIP2023-112, pp. 95-96, 2024年2月. |
資料番号 |
HIP2023-112 |
発行日 |
2024-02-15 (HIP) |
ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
HIP2023-112 |