Due to the advanced image processing and the development of network technology, high-definition video capturing equipment has become common, and videos are uploaded on various platforms. As one of them, studies to provide FVV(Free Viewpoint Video) services are actively being conducted. FVV service is one of immersive media services that composes and provides videos of various views in one content so that a user can enjoy it at a desired location or viewpoint.
However, despite long-term research and discussion of standardization for FVV services, several problems to be solved for commercialization remain. In particular, in terms of video transmission, it is physically very difficult to simultaneously transmit all the view video and additional data constituting FVV.
For FVV to be streamed, additional consideration for FVV is required. FVV must cope with problems that may arise while providing multiple videos as well as transmission problems caused by changes in network environments. Since the FVV service cannot provide all the videos in one content by streaming at once, it is necessary to select which video to provide first. To increase the number of videos to be transmitted, lowering the quality of the video and providing it degrades the overall QoS(Quality of Service). On the other hand, the method of frequently changing video causes requests and transmission of large-capacity media, which can lead to overload of the server, resulting in delays or errors in service. After all, in order to stream a large-capacity video service such as FVV, it is important to select and transmit an appropriate video.
This paper proposes an object-centric dynamic adaptive streaming method for FVV services. A method of searching for an optimal recognition scene for each object in video content photographing multiple viewpoints and configuring information on the corresponding sections is proposed. A video list file can be configured so that each object can be viewed through the configured information. Through the provided information, each client can request a video related to an object that the user wants to watch. In addition, in order to maximize the video service enjoyed based on the object, an object-centric zoom in/out screen streaming method in FVV is additionally proposed.
The proposed method shows improved performance compared to the existing method in three aspects. First, it reduced the video search time. The proposed method does not require a separate search after initial acquisition of video information when viewing an object-centric FVV. Second, it reduced the amount of video transmission. Third, the probability that the object can be viewed optimally is increased. Since the proposed method has already searched for a video section with a high recognition rate through object recognition, the best object-centric screen can be guaranteed in most sections.