Inspiration
"A picture is worth a thousand words" - The limitless possibilities of extracting knowledge from TV content, which heavily influences the daily lives of nearly all germans
What it does
Our ML algorithms can distinguish between the logos of all the given german TV channels. But it doesn't stop there: we also decided to bring in the capabilities of the Azure Cloud to gain further knowledge about the channels and their content by means of analyzing the faces in their shows. Thereby, we could find out about recurring people, the emotions mainly displayed, the M/F ratio, the average age and so much more of each station and gain some interesting insights.
How we built it
As a team :)
- Azure CentOS custom image
- various Azure Computer Vision APIs
- standard python data science stack
- NumPy
- Pandas
- OpenCV
- Matplotlib
- Jupyter
Challenges we ran into
With the logo detection:
- We had a limited amount of poorly labeled data.
- On some recorded frames, logos were hardly visible.
- On some recorded frames, parts of a title or something alike would be similar to logos, so they might be mislabeled
Accomplishments that we're proud of
- Fast and accurate logo detection
- Lots of new data gained from the face recognition, found out a bunch of fun facts
What we learned
- We learnt to use Microsoft Azure Cognitive Services, including:
- Azure Face API
- Azure Emotion API
- Azure Computer Vision API
- Some OpenCV
What's next for logo404
Given more data, we can perform extensive data analysis. For example, we might automate the labeling of the actors appearing in each show. On top of that, there's tons of possible usecases with the insights gained from the analysis: -in combination with the viewer demographics, we might find better suggestions for which shows to display on each channel -depending on the emotions being shown in a transmission, we might recommend possible categories of adds to show alongside (as people's buying decisions highly depend on their emotions)
Log in or sign up for Devpost to join the conversation.