Real-time feature extraction from video stream data for stream ...

Real-time feature extraction from video stream data for stream ...


Learning tasks

This thesis is about the identification, extraction and evaluation of features from video

data. Video data gets produced in countless settings, including television broadcasts,

Internet videos, and surveillance cameras. As the results of this thesis are supposed to

be used within the ViSTA-TV project, I decided to base the biggest part of my thesis

on the analyze of video features, gained from IP-TV data. Nevertheless, the features

extraction process and the evaluation of the extracted features are similar for other

settings as well. Hence, I decided to build up another use case, that is closer related

to surveillance than television. For both use cases, the corresponding learning tasks are

defined in this chapter.

4.1. Use case A: IP-TV

As mentioned in the introduction, one important task of the ViSTA-TV project is to

understand the interests and behavior of IP-TV users. This shall then be used to predict

user switches. To solve this task, it is necessary to have detailed information about the

broadcasted television program. Some information about the program already comes

with the EPG data, which is provided by the broadcasting companies. We are able to

receive the EPG data through our ViSTA-TV project partner Zattoo. Unfortunately the

EPG data is limited to information on the show level, including

start and end time of a broadcast, accurate to the nearest second,

title of the show. In case the show is part of a series, the episode title is also given,

categories the show falls into. This includes for example documentation, news, movie,

sports, ...

Hence the segmentation and tagging of the incoming video stream on the show level is

already given. But in order to gain better predictions on user behavior and interests, we

are interested in a closer grained segmentation. Thus, I decided to take the segmentation


More magazines by this user
Similar magazines