We always had an interest in video and filmmaking. Analysing frames was the perfect fit for us.
It uses machine learning or opencv to detect a logo in the whole frame, and identifies it.
We started with computer vision api from microsoft. Once that didn't worked for our needs and after testing other Azure services we decided to use scikitlearn and opencv. With those services, we were able to detect and identify a logo, given that the position of the logo was known. At the end we got the best results with the opencv template matching, where we are able to find the logo in the whole place, with a high accuracy.
Finding and installing all the right tools and apis took so much time, that it meant we only had less 24 hours to implement it. The proposed tools by microsoft, while being interesting, did not fulfil our requirements.
We did not really expect to have a running program at the end, that was able to detect logos in the whole image.
We got to get know a lot of tools that are used for machine learning. Especially those which are centred around image recognition.
We would like to make it possible to detect illegal video content, by recognising logos in low quality frames.
Log in or sign up for Devpost to join the conversation.