二つのPTZカメラを用いて移動物体を追跡と三次元計測システムの開発

カテゴリ: 研究会(論文単位)

論文No: IP10006

グループ名: 【C】電子・情報・システム部門 情報処理研究会

発行日: 2010/03/26

タイトル(英語): Moving Object Tracking and 3D Measurement Using Two PTZ Cameras

著者名: 鄭 光(徳島大学),大恵 俊一郎(四国大学)

著者名(英語): Zheng Guang(The University of Tokushima),Oe Shunichiro(Shikoku University)

キーワード: トラッキング|パーティクルフィルタ|ステレオ|ＳＩＦＴ特徴|三次元計測|Visual Tracking|Particle Filter|Stereo Method|SIFT Features|3D Measurement

要約(日本語): This paper proposes a new method to track and measure 3D informationof moving object by using two PTZ(Pan-tilt Zoom) cameras. The traditional stereovision is mainly based on static cameras, the optical axes of twocameras are required to be parallel or a fixed angle, the 3Dinformation is calculated by the disparity of two cameras. In our 3Dmeasurement method, the optical axis of two PTZ cameras are notrequired to be parallel or a fixed angle, the 3D information can bemeasured even though the pan and tilt of two PTZ cameras are changedfreely. The SIFT features are detected in the tracking result of two PTZcameras, due to SIFT feature is invariant to rotation, translationand scale variation, the corresponding points in the left and rightimage can be determined well. Therefore, our method can track theobject and the 3Dinformation of the moving object can be calculated simultaneity. The results ofexperiment performed in this study show the effect of our approach.

要約(英語): This paper proposes a new method to track and measure 3D informationof moving object by using two PTZ(Pan-tilt Zoom) cameras. The traditional stereovision is mainly based on static cameras, the optical axes of twocameras are required to be parallel or a fixed angle, the 3Dinformation is calculated by the disparity of two cameras. In our 3Dmeasurement method, the optical axis of two PTZ cameras are notrequired to be parallel or a fixed angle, the 3D information can bemeasured even though the pan and tilt of two PTZ cameras are changedfreely. The SIFT features are detected in the tracking result of two PTZcameras, due to SIFT feature is invariant to rotation, translationand scale variation, the corresponding points in the left and rightimage can be determined well. Therefore, our method can track theobject and the 3Dinformation of the moving object can be calculated simultaneity. The results ofexperiment performed in this study show the effect of our approach.

原稿種別: 日本語

PDFファイルサイズ: 9,948 Kバイト