Good analysis. The biggest problem with machine vision approaches (in addition apparently to the fact that the machine vision part doesn't seem to work very well) is that the technology still can't understand what it is looking at. It may know two frames are the *same* but still doesn't know who or what is in them. Most approaches I've seen to this are then to have humans tag the frames, but then you are back to the same problem any editorial approach has with respect to search-it simply can't scale with the volume of content.
Christophor Rick
· 1 year ago
Hello Tom,
Thanks for the comments. I should think that it would be easy enough to make a computer understand the human face. After all, generally speaking, they are all the same. Two eyes within a certain distance, a nose, a mouth. Mostly eyebrows etc... If someone is going to do facial analysis I should think they could do it with a broad template and be able to match it then. Additionally another way I could see it being done was to have representative images of specific people and match the frame in the video against the image allowing for a small margin of error to take into account angle and such. It's an interesting field and one the government is, I'm sure, extremely interested in so they can process millions of minutes of video footage from public cameras to find the evildoers.
Suddenly I have this sneaking suspicion that we're like in 1984 and we just don't know it.
Anyway, I'm curious to see how it progresses and if it becomes effective
All Comments
Good analysis. The biggest problem with machine vision approaches (in addition apparently to the fact that the machine vision part doesn't seem to work very well) is that the technology still can't understand what it is looking at. It may know two frames are the *same* but still doesn't know who or what is in them. Most approaches I've seen to this are then to have humans tag the frames, but then you are back to the same problem any editorial approach has with respect to search-it simply can't scale with the volume of content.
Thanks for the comments. I should think that it would be easy enough to make a computer understand the human face. After all, generally speaking, they are all the same. Two eyes within a certain distance, a nose, a mouth. Mostly eyebrows etc... If someone is going to do facial analysis I should think they could do it with a broad template and be able to match it then. Additionally another way I could see it being done was to have representative images of specific people and match the frame in the video against the image allowing for a small margin of error to take into account angle and such. It's an interesting field and one the government is, I'm sure, extremely interested in so they can process millions of minutes of video footage from public cameras to find the evildoers.
Suddenly I have this sneaking suspicion that we're like in 1984 and we just don't know it.
Anyway, I'm curious to see how it progresses and if it becomes effective