The Study on Moving Object Detection Method Based on Aerial Video

• Main Authors: Zai-Ci Jiao, 焦在麒
• Other Authors: Tsong-Yi Chen
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/bexx6k

碩士 === 國立高雄應用科技大學 === 電子工程系碩士班 === 102 === The project proposes a moving object detection method used for the aerial video, the system is mainly composed of three parts: (1) Classification of feature points: It is first to search the feature points at each frame by use of motion vector and label the high-density distribution area of feature points. By the Delaunay triangulation technique, each feature point is classified into a moving-object area or the background. (2) Generation of moving-object mask: Based on the above classification result of feature points, the motion vectors of feature points on background are employed for deriving a conversion matrix by perspective projection. Thus, such a conversion matrix is used to obtain deformation of background in the neighboring frame and the background deformation is used to form the moving-object’s contour by frame-difference using the current frame. Finally, a moving-object mask is generated by utilizing the moving-object’s contour with the morphological processing. (3) Labeling & tracking of the moving-object: From the above masks of moving-objects, each moving-object is then labeled and tracked. The early experimental results show that the proposed method can real-time detect & track moving-objects on the ground effectively. The developed technique can be applied to the aerial vehicle-tracking system, the aerial search system, and the visual system of the unmanned flight vehicle.