How to detect several objects and their features from a Video? [closed]
I am building a video search engine. Its main task is to search the videos from the video dataset similar to the video provided by the user. I have tried it in two different ways, first by HISTOGRAM difference and second by using SSIM algorithm to compare each key-frame of both the videos in comparison. But now I want to try a different approach. I want to detect various objects from the various videos from my training video set and store their properties, like shape,size,color,texture,entropy,frequency,etc. now using this objects and their features I can categorize them. Thus which would reduce the number of comparison with the various videos. The video uploaded by the user will be compared with the videos having similar objects and similar features.
But I am confused how to do it. I tried reading about cascading files but couldn't get any hint. So, please if anyone of you know about this or have any ideas can ping me at [email protected]
it seems, you underestimate the costs / difficulty of object recognition. none of the ideas you mention are even vaguely suitable.
seriously, this will take you some more years
(again, i'm not trying to putyou down, but .. this would need much more effort (andskill), than you've shown so far)
It will be much easier to directly compare frames until you reach a good enough degree of confidence that it is in fact the same video you have in your database. As berak said, your idea of recognizing objects to match videos is like solving the AI problem altogether :) don't go there.
@Johnson, it's unclear, what your real requirement is. "similarity between videos" and "object detection" might be 2 very different pair of shoes
@berak using the various objects in a particular video to compare with the objects of other video for example : if there is a particular person in one video then searching a similar video having that specific person
@pedro-batista Thanks for the suggestion, but comparing videos frame by frame just takes too much time
@berak if object-feature comparison is not feasible can you suggest other methods to compare two videos and check similarity between them
And you think detecting particular objects in a stream of images does not take time?