| Title | : | Video Stabilization and Moving Object Segmentation in Jittery Videos |
| Speaker | : | Geethu Miriam Jacob (IITM) |
| Details | : | Thu, 20 Oct, 2016 3:00 PM @ BSB 361 |
| Abstract: | : | Videos have become an integral part of a fast growing digital world. There has been much advancements of research in the field of video analytics. The common tasks in this area include Video Summarization, Video Retrieval, Video Stabilization, Video Compression, Motion Segmentation, Tracking, Crowd Analysis, Anomaly detection, Activity Analysis, Video Description, etc.
Moving object segmentation is the process of obtaining the silhouette of the moving object from a video. This task is challenging when the video is jittery. Frameworks for video stabilization and moving object segmentation in jittery videos have been proposed. Video Stabilization is performed by Procrustes analysis of the trajectories. The trajectories belonging to the background are modeled in Kendall’s shape space, thus estimating the jittery camera motion. This camera motion is stabilized to render stable frames. Also, stabilization of trajectories in jittery videos improves the separability of the trajectories of the moving object and the background. This helps in tracking the moving object in the jittery video accurately, which in turn reduces the segmentation error of the moving object. Performance analysis with respect to recent state-of-the-art methods exhibits the superiority of our proposed methods. The future scope of our work will target methods for video captioning. Video captioning, a process of converting videos to sentences, is a high level task in the field of video analytics. The seminar concludes with an overview of the existing methods for video captioning. |
