Video Advertising: More Intelligent With Computer Vision Technology

Online video advertising combines a video and an advertising image/video

Nowadays almost all online ads are based on keywords extracted from text of a page, finding the most relevant ads and putting them up alongside the page, such as AdWords by Google. Can we have “Video AdWords” extracted directly from online videos?


Video advertising is to generate contextual ads by analyzing the content of online videos. It has wide applications in non-intrusive online video advertising (e.g. YouTube) and TV program advertising (e.g. ESPN sports or Movies). The visual/text ads can be embedded into the flat/blank areas (usually non-distracting areas, such as sports court, ground, wall, etc.) or put in the corner of a video, for some duration of trivial moment (e.g., during short intermission). 

Embed ads into sports video

Embed ads into sports video (From PVi)

Computer vision technology makes video advertising more intelligent. It could automatically decide which ads to select, when and where to embed the ads, and how to synthesize a new video for the selected ads and given video.

Video advertising using computer vision technology usually consists two steps:

  1. Video content analysis (Video analytics): to recognize objects or people in online videos and automatically generate contextual ads. The challenge in this step is how to learn the best context concept from video clips (not text).
  2. Embed ads into video: to automatically embed the generated contextual ads into online videos. This step is similar to what we discussed in a previous post: Embed Visual Ads into Videos. The challenge in this step lies in 3D perspective estimation, lighting estimation and photorealistic rendering of the new video with correct shadow. For example, the following figure from PVi shows shadow inconsistency of ads object and players in court. This could be solved by rendering the scene using captured environmental lighting or estimated lighting condition. 

Inconsistency of shadow directions of ads object and player

I believe video advertising will be the next wave of online advertising in the next few years. The market can be imagined from the rising of YouTube, which has advanced to the second largest search engine after Google. Many start-ups are targeting at this market. We will discuss more companies and technologies on video advertising in future posts.

1 Comment

Other Links to this Post

  1. VisionWang: Computer Vision Expert Blog » Video Advertising: More … « provaaa — January 3, 2009 @ 7:29 am

RSS feed for comments on this post. TrackBack URI

Leave a comment