Interactive Broadcast Services For Live Soccer Video Based On Instant Semantics Acquisition

摘要

This paper presents a system for providing interactive broadcast services for live soccer video that is based on instant semantics acquisition. Currently, we have implemented two such interactive services: live event alert and on-the-fly language selection. The live event alert service has a small time lag of about 30 s for a short video clip to reach its final viewer and at most 1.5 min for a long clip of the live event. The on-the-fly language selection service allows users to choose their preferred contents and preferred language. The motivation for this work is that such interactive services will greatly increase the value of live soccer video. Currently, similar systems attempt to derive semantics of a soccer game from gamelog in freestyle text format and low-level features of the video, which is a challenging task. In this paper, we tackle this challenge with a combination of both gamelog input tool and targeted algorithm proposed in this paper. Our system is powered by our proposed semantic gamelog input tool that facilitates fast and accurate input of a semantic gamelog that contains basic semantic information of atomic events. When an interesting event occurs, our system performs boundary detection of these events by combining features extracted from the video with additional information from the semantic gamelog. This additional information facilitates our system to achieve accurate and very fast boundary detection of these events to support our live event alert service. Our system also implements a gamelog translation machine which translates the semantic gamelog (encoded in a game-specific code) into any natural language, provided that there is a configuration file for that language. Combining our gamelog translation machine with existing text-to-speech technology, we provide the on-the-fly language selection service. (Currently, our system supports English, Chinese, and Malay.)