Update 2025-02-13-logo-detection.md

Author: danmalter

_posts/2025-02-13-logo-detection.md

@@ -13,7 +13,7 @@ featured: false
 
 In the world of sports, sponsorship and advertising are integral to a team or league’s revenue. Companies invest millions of dollars to place their logos in key spots for TV viewers. For example, along the boards of the ice during an NHL game or rotating sideboards of MLB, NBA and Premier League soccer games. But how is the return on investment of these sponsorships measured? It’s well known how TV viewership varies throughout the broadcasts of each sport and game time slot, but quantifying the exact exposure - and thus value - of an advertisement remains a challenge. Using TV viewership data along with an AI logo detection model, we can solve this challenge with a new business metric in sports advertising; Total Viewership per Second of Exposure. For the purposes of this demonstration, we are using the generic [Google AI Logo Detection]( https://cloud.google.com/vision/docs/detecting-logos){:target="_blank"} model but this or similar models can be retrained to learn better about identifying certain logos. 
 
-Using Python, we used a video clip from the first period of a game between the [Chicago Blackhawks and St. Louis Blues]( https://www.youtube.com/watch?v=_P9cXD2BqFU){:target="_blank"}, played on February 8th, 2025 at the Enterprise Center, broadcasted by Comcast Sports Network. Hockey is a fast-paced sport, so it’s unlikely that a logo will be 100% visible for more than a few seconds at a time. To solve this, we’ve written an algorithm that uses the Google model to first identify if a logo is in a given frame. We continue to track that logo over the next two frames to detect if the logo remains visible; if we capture 100% of a logo in three consecutive frames, we count the occurrence as one continuous logo detection. We ran a highlight video from the first quarter of this game and below is the total amount of seconds that each logo was shown over a 93 second video. 
+Using Python, we used a video clip from the first period of a game between the [Chicago Blackhawks and St. Louis Blues]( https://www.youtube.com/watch?v=_P9cXD2BqFU){:target="_blank"}, played on February 8th, 2025 at the Enterprise Center, broadcasted by Comcast Sports Network. Hockey is a fast-paced sport, so it’s unlikely that a logo will be 100% visible for more than a few seconds at a time. To solve this, we’ve written an algorithm that uses the Google model to first identify if a logo is in a given frame. We continue to track that logo over the next two frames to detect if the logo remains visible; if we capture 100% of a logo in three consecutive frames, we count the occurrence as one continuous logo detection. We ran a highlight video from the first period of this game and below is the total amount of seconds that each logo was shown over a 93 second video. 
 
 <div style="text-align: center;">
   <video width="640" controls>