Descripció del projecte
The rapid growth of the Internet, and the incredible flow of information that the Internet has made possible, has transformed the business of advertising. Today it is difficult to surf the web without seeing online advertising, often in the form of visual display ads on web sites (including pop-ups and pop-downs) and textual ads on search sites. There is little doubt that on-line advertising has taken business away from traditional modes of advertising, such as newspapers, snail mail, and radio. Internet advertising is key for the business models of lots of internet video portals.
If we have a closer look to the digital advertising methods, online video ads are one of the fastest-growing ad mediums, far outpacing growth in spending on television and other digital formats. Online video ad viewing exploded in 2013 and is growing faster than most other advertising formats and mediums. Video ads provide a level of visual and narrative richness that nearly equals television, while offering all the advantages of digital, including advanced targeting, tracking, and increasingly, automated buying of video ad units. Moreover, mobile video consumption is growing rapidly and providing advertisers with a way to reach consumers when they are paying attention. Between 2012 and 2014, smartphone and tablet video consumption grew 400 percent and now accounts for 30 percent of all online videos played. Today, the largest advertising performance is achieved through in-video mobile advertising.
The best video advertising is watched by Web users and not perceived as disruptive, but creating that type of content can be a challenge. For years, content publishers have been struggling to establish best practices for deploying ads in video content. As more and more video content becomes available, online marketers have been testing various strategies to see what consumers will tolerate and where ads can deliver strong results. A big point of concern is ad placement. Ads can be placed before, after or in the middle of a piece of video content. In this project we want to go and step forward in ad video placement.
The main objectives of the thesis project are
(1) To automatically detect moments of the video where to place the ads. We define these moments as famous people which is the image of a brand (character or celebrities detection)– a scene with George Clooney can be placed with a Nespresso advert.
(2) To automatically detect the emotion of the faces/celebrities detected. So that best adverts can be placed with positive sentiments of the video scenes.
Several research and development areas are involved in the whole process:
– Computer vision techniques will play a very important role in the project. Face recognition, scene and object classification will be used to detect the stellar moments. Moreover several of these techniques can also be used to insert the advert in the video in a more natural way, taking into account features such as lighting conditions, and so on.
– NLP. Automatically labelling appearances of characters in TV or film material with their names is tremendously challenging due to the huge variation in imaged appearance of each character and the weakness and ambiguity of available annotation. However, high precision can be achieved by combining multiple sources of information, both visual and textual. NLP over subtitles and transcripts will be used to complement visual cues for detecting stellar moments.
– Audio. Audio algorithms can be used together with computer vision and NLP to detect stellar moments. Low-level processing such as intensity variation can be used to detect action vs romantic scenes. Other more advanced techniques such as semantic meaning can be used to identify what the characters are talking about (speech to text).
– Machine learning over the detected moments will be used to learn which are the adverts that best fit the video in terms of brands interests.
– Data mining will be used to build user behavioral models and personalize the adverts to deliver.
