China 
Africa 
Explore chapters and articles related to this topic
Digital video processing
Published in John Watkinson, An Introduction to Digital Video, 2012
Figure 4.29(a) shows an example of a moving object which is in a different place in each of three pictures. The optic flow axis is shown. The object is not moving with respect to the optic flow axis and if this axis can be found some very useful results are obtained. The process of finding the optic flow axis is called motion estimation. Motion estimation is literally a process which analyses successive pictures and determines how objects move from one to the next. It is an important enabling technology because of the way it parallels the action of the human eye.
Image portrayal
Published in John Watkinson, Convergence in Broadcast and Communications Media, 2001
Figure 7.50(a) shows an example of a moving object which is in a different place in each of three pictures. The optic flow axis is shown. The object is not moving with respect to the optic flow axis and if this axis can be found some very useful results are obtained. The proces of finding the optic flow axis is called motion estimation. Motion estimation is literally a process which analyses successive pictures and determines how objects move from one to the next. It is an important enabling technology because of the way it parallels the action of the human eye.
Digital Video processing
Published in John Watkinson, The Art of Digital Video, 2013
Figure 5.43a shows an example of a moving object that is in a different place in each of three pictures. The optic flow axis is shown. The object is not moving with respect to the optic flow axis and if this axis can be found some very useful results are obtained. The process of finding the optic flow axis is called motion estimation. Motion estimation is literally a process that analyses successive pictures and determines how objects move from one to the next. It is an important enabling technology because of the way it parallels the action of the human eye.
A video painterly stylization using semantic segmentation
Published in Journal of the Chinese Institute of Engineers, 2022
Der-Lor Way, Rong-Jie Chang, Chin-Chen Chang, Zen-Chung Shih
Temporal coherence is crucial in a stylized video; a video has severe flickering without coherence. Optical flow is the most commonly used motion estimation method for achieving temporal coherence. However, it is limited in scenarios with large displacements and inaccurate motion boundaries. Semantic segmentation can enhance optical flow by providing information about object locations and boundaries to address these inaccurate motion boundaries. Therefore, this paper provides a semantic segmentation–based style transfer method for videos. After inputting a video and two style images, our proposed method segments both foreground and background objects and transfers their styles. This approach can overcome the problems of incorrect motion boundaries, disjointed segmentations, and low-level noise in stylized images.
Self-supervised optical flow derotation network for rotation estimation of a spherical camera
Published in Advanced Robotics, 2021
Dabae Kim, Sarthak Pathak, Alessandro Moro, Atsushi Yamashita, Hajime Asama
Motion estimation of cameras is essential in robotic applications such as simultaneous localization and mapping (SLAM) [1] and structure from motion [2] as these applications require the movement of cameras. Recently, learning-based approaches have adopted for camera motion estimation using convolutional neural networks (CNNs) [3–5], and these approaches have performed equivalently or better than the conventional feature-based approaches. Among them, fully supervised learning approaches have been often utilized to regress the camera motion using raw images or optical flow fields as inputs. However, they require a large number of datasets with accurate labels, which are difficult to acquire. Many attempts have been made to capture such datasets using motion capture systems [6,7], GPS/IMU systems [8,9] and other sensors [10]. However, they often require precise sensor calibrations and collection/annotation expenses. Meanwhile, self-supervised learning approaches have recently emerged for scenarios in which it is difficult to acquire such labels, e.g. in the field of person re-identification [11], video hashing [12], and image classification [13]. These approaches offer the benefit that they do not require any labeled data, as they only utilize the collected data. For the camera motion estimation, many self-supervised approaches have been employed [14–17] and have shown estimation results comparable or better than fully supervised learning approaches.