드론 동영상을 통한 광학 흐름 기반 프레임 보간 모델 성능 분석

다국어 초록 (Multilingual Abstract)

Optical flow-based video frame interpolation is a technology that receives an actual frame as input and generates a frame corresponding to an arbitrary time between the frames through the estimated optical flow. This video frame interpolation method was mainly conducted on images with only flat movements, such as Vimeo90k and DAVIS, where altitude information of photographed cameras was not largely considered, and the optical flow at this time shows a unidirectional movement. However, in videos in which the distance to the subject changes depending on the altitude of the camera being photographed, such as drone images, the optical flow is expected to converge and diverge as well as unidirectional movement, and experiments to analyze the performance of frame interpolation in this optical flow are insufficient. Therefore, this paper aims to analyze the performance of optical flow-based frame interpolation method through videos showing optical flow in the form of convergence and divergence according to the altitude information of the camera such as drone images.