inVidAd

Inspiration

Ads in videos are boring. Everybody hates ads popping up in the middle of your video! Most humans give close to no attention to products advertised! On the other hand, smaller content creators do not have traction and viewership metrics. Two address these two problems, we present inVidAd, where we place ads strategically and adaptively in your video.

What it does

-We identify AdSpaces in your video and optimally place Ads based on your geography and sentiment/mood, as observed in the past. -We insert specific Ads in regions of interest (RoIs) we identify. -We adapt it to the region and sentiment of the end-user. This means that Starbucks coffee mug Ads may not sell in South Asia. -This acts as a new revenue stream to the creator, new scope of user attention for small business advertised and redefines how we look at Ads as a whole

How we built it

To detect Ad space, we used multiple techniques like image segmentation, and boundary detection using contour and foreground-background models. Once the Ad space is identified, we infuse Ads into the video frame by frame. We also ended up exploiting grayscale features of the core video and used the best RoIs we got from this analytical framework to place the Ad.

Challenges we ran into

Our approach works best for stable videos like podcasts. -Detecting Stable Ad space in fast-paced videos.
-Occlusion detection and removing Ad where appropriate -Product Market Fit -User behaviour for Ads -User sentiment research

Accomplishments that we're proud of

Adaptively being able to infuse both videos and images into dynamic AdSpaces.
Occlusion detection
Building a fantastic end-to-end system with high detection accuracy
Helping small businesses promote their products.

What we learned

Object detection using OpenCV Foreground and Background detection using YOLO7 React.js Django

What's next for inVidAd

To increase the area of AdSpaces, we can also detect textures and patterns in the foreground. We can further stabilize ad placement by addressing the problem of occlusion while detecting spaces. We can also adaptively show ads based on past user history of the user.