Topic 5: Jiku: Live (Video) Event Capturing and Sharing

Large dense cities often hold large-scale outdoor events (e.g., parades, races, rallies, art performances). These events are usually crowded, and may spread across a large area. Thus, most attendees do not have a good and complete view of the event. To improve the event experience, we are developing a system called Jiku (“instantly cool” in Chinese) where attendees can access, using their mobile phones, video streams of the events. Figure 5.1 illustrates the architecture of Jiku. The video streams can be captured using video cameras provided by the organizers or mobile cameras of other attendees. Using the Jiku mobile app, the attendees can browse, watch, interact, record, and share video streams of the events, providing a personalized experience in which attendees can watch the part of the events that is interesting to them, possibly from a different angle from where their physical location.

Besides supporting browsing, the Jiku mobile app uses the choices (both real-time and historical) from other attendees to rank the videos, and the most interesting ones are recommended to the attendees. While watching the video stream, the attendee can interact with the stream, using multi-touch gestures to zoom and pan within the video. The recommended regions of interests are hinted to the attendees so that they can quickly zoom into the interesting regions and track the object of interests in the video.  Alternatively, the attendee can choose to subscribe to an automatically generated stream that switches between the existing video cameras based on the visual quality of the streams and cinematography rules.

The attendee can choose to record and share the video streams they like with either other attendees or with others who are not attending the event.  Recording is done through the cameras on the mobile device.  Jiku employs an efficient video enhancement algorithm that improves the quality of the videos captured under challenging light conditions.

Figure 5.1: Architecture of Jiku. 

Sharing is challenging since at such crowded event, the mobile network infrastructure is usually saturated and does not have enough uplink bandwidth to the mobile clients to upload the video directly. We address this challenge and achieve sharing in several ways. If the video to be shared originated from the organizer’s cameras, the video is recorded on the server and sent to the social sharing website directly, bypassing the mobile upload process.  If the video to be shared originated from a mobile camera, then the video is either transmitted through a mobile-to-mobile connection directly to other attendees, or uploaded opportunistically when the network connection allows. The upload scheduler decides which video has the upload priority by selecting the subset of all recorded videos with the best coverage of the event. Jiku also implements a mobile-to-mobile video recommendation system that propagates rating information among the mobile phones for a distributed collaborative filtering algorithm to compute which videos should be recommended to which users.

Jiku is supported by a protocol stack that improves on the reliability and throughput of wireless transmission. Jiku exploits multiple network interfaces to simultaneously transmit the video, uses a modified TCP congestion control algorithm that is optimized for long bandwidth-delay link, and employs error-estimation coding to adapt the wireless data rate and to retransmit video packets in a BER-aware manner. Further, Jiku allocates the unicast and multicast channels appropriately to stream the region-of-interests to the attendees, in a way that increases the number of attendees supported without overwhelming the recipients with unwanted multicast packets.

Representative Publication

[1] Arash Shafiei, Ravindra Guntur, Ngo Quang Minh Khiem, Wei Tsang Ooi: Jiku Live: Live Zoomable Video Streaming. ACM International Conference on Multimedia (MM), 29 October - 2 November 2012, Nara, Japan.

[2] Mukesh Kumar Saini, Raghudeep Gadde, Shuicheng Yan, Wei Tsang Ooi: MoViMash: Online Mobile Video Mashup. ACM International Conference on Multimedia (MM), 29 October - 2 November 2012, Nara, Japan.

[3] Binbin Chen, Ziling Zhou, Yuda Zhao, and Haifeng Yu. Efficient error estimating coding: feasibility and applications. SIGCOMM Comput. Commun. Review. 41, 4 (August 2010)

[4] Jingyuan Wang, Jiangtao Wen, Jun Zhang and Yuxing Han: TCP-FIT - an improved TCP congestion control algorithm and its performance. IEEE INFOCOM 2011, Shanghai, October, 2011.

[5] Padmanabha Venkatagiri Seshadri, Mun Choon Chan, Wei Tsang Ooi: Mobile-to-Mobile Video Recommendation. 9th International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services, 12-14 December 2012, Beijing, China.