Need:
- With the ever-growth large-scale video in the mobile phone, so what will everyone get from these video? There are many videos contain something very interesting like a short comedy video. So if someone find something interesting in the video and want know more about it, they may not search it in the internet and find the information after watching this video due the poor memory. So if the advertiser have put some advertisements in the video ahead of time, it will be more convenient for the user to get some information. That’s very useful for the advertisers and the users.
- There are many videos in users’ phone. Maybe most of them are meaningful time mark. So someone want to look for some useful tools to tagging the meaningful object or want to know the object information. Then our video tagging systems will be very efficient for this work.
Approach:
- The video tagging project can be divided into two steps. The first one is the key frame localization. The second one is the object classification or object detection.
- The key frame localization can be realized by some conventional method like the HOG features split or some other method. This is a litter challenge because there is no very efficient way to get the really accuracy key frame. And I think it is a program optimization problem.
- The object classification can be realized by the deep convolutional neural network classifier or some other deep learning state-of-the-arts method. The problem is the labels may be not enough. So it can be a research problem.
Benefit:
- Everyone can be convenient to get some merchandise information by the tagged video which is processed by the mobile end application.
- Some people will summarize the meaningful moments and find some meaningful object.
Competitors:
There a video tagging system which has been released in the internet after my survey. The Website name is “Clarifai”. They can tag the video and get the object temporal information. And the classification accuracy is very high. So it is our main competitor.
10/18/2015
Fuchen Long