1. Wide Area High Definition Video Streaming for Tiled Displays Duy-Quoc Lai, Falko Kuester, Stephen Jenks, Zhiyu He Laid@uci.edu · www.duyonline.com · www.research.calit2.net/students/surf-it2006 · www.calit2.net S ummer U ndergraduate 2 R esearch 0 F ellowship in 0 I nformation 6 T echnology Introduction High-definition video streaming over wide area networks and display on tiled displays presents several research problems. The high-definition video has to be captured, streamed and displayed at interactive rates, while being subjected to bandwidth and latency limitations. Since the size of an HD stream will exceed the capabilities of commodity gigabit interconnects, data compression is required. Furthermore, it has to be possible to deliver the appropriate portions of the video frames to the corresponding tiles of the display, while retaining the ability to freely and smoothly scale and move the video across the display wall. The project goal is to demonstrate the feasibility of high-definition video streaming in support of distributed collaborative digital workspaces. High Definition Video Standards Video capturing is done with a Panasonic AG-HVX200 camera. The camera is capable of capturing video at various formats. Fig. 1: Live video streaming at 1920 x 1080i p30. Shows the quality that can be captured with a HD camera. By experimenting with the video format, limiting the amount of data being sent over the network, and capping the number of frames per second, the network is observed to be the main bottleneck; after which comes the texture mapping. At high resolution, this limitation becomes obvious: only a few frames can get through the network every second. This leads to having only a few frames to texture map, resulting in a low frame rate. It is difficult to achieve 30fps 1080i 30p HD video streaming, but it can be done with compression and optimization. Results Acknowledgement We would like to thank CAL-IT 2 and CAL-IT 2 SURF-IT Fellowship program for providing support for this research. We also thank Harry Mangalam for his support on this project. Fig. 2: The interlacing effect is shown from capturing video using 1080i 60i format. 1080i 30p format is preferred because it captures the entire frame without interlace. Video Streaming Pipeline Fig. 3: A live video stream is generated from the HD camera and goes through a series of processes before it can be displayed on the HIPerWall. Fig. 4: There is a time delay from when an event is captured on the camera until the event is displayed on the HIPerWall with respect to the resolution. Fig. 5: The number of frames that gets texture mapped and displayed per second. FormatResolution 1080i 60i1920 x 1080 60i interlace (highest resolution for today’s standards). 1080i 30p1920 x 1080 30p progressive 720p1280 x 720 progressive Fig. 6: Amount of data sent across the network with respect to the resolution. Notice: The network caps out at 62MB/second. The project is divided in 3 areas: image decompression, network transfer, and texture mapping. Image Decompression The video stream sent from the camera is decompressed into a format that OpenGL supports. Network Transfer Uncompressed bitmap of each frame is sent over the network to the group of display nodes. Texture Mapping Each display node takes the bitmap and texture map it on to a 2D plane.