Mobile Intelligence

21.5 MoVR: MOBILE-BASED VIDEO RETRIEVAL

In this chapter, the MoVR framework is proposed for the mobile-based video retrieval system development. This framework can support not only basic event queries but also the complicated queries toward some temporal event pattern, which consists of a set of important events followed by a certain temporal sequence. More importantly, this framework is capable of providing both personalized recommendations and generalized recommendations. Users can also specify a fuzzy weight parameter if their query interests have not yet been clearly formed. The system will make the adjustment and generate different retrieval results based on the fuzzy-associated queries. In essence, MoVR is designed to provide not only powerful retrieval capabilities but also a portable and flexible solution to the mobile users.

21.5.1 Overall Framework of MoVR

As shown in Figure 21.3, the overall framework of MoVR includes three main processing phases on the server side.

• Phase 1 is for video data preprocessing. It consists of the following steps. The first step is to process the source video data to detect the video shot boundaries, segment the video, and extract the shot-based features. Data cleaning and event annotation algorithms are then applied to detect the anticipated semantic events by employing the extracted shot-level features and multimodal data mining scheme. The components in Phase 1 are processed offline, which is not the focus of this chapter.
• Phase 2 is to ...