Any Software That Do Video Analytics Of Stored Video?

We have more than 100 TB of stored video and want to do some video analitics of the videos

and produce clips smaller in size and easier to store

Any suggestions?


What kind of analytics?

Have you looked at Briefcam?

If the video is not already tagged with meta-data (and it most likely isn't), it's going to be VERY time-consuming to do anything with stored video. As a rule of thumb, you're looking at a processing time of about 2x-5x the during of stored video. EG: if you have 100 hours of video, it could take anywhere from 200 to 500 hours to process that into something that has a degree of intelligence applied to it.

But I'm not clear from your question what exactly you want to achieve. If you just want smaller size clips/storage, have you considered reencoding to lower resolution or frame rate?

Nario, my questions are similar to undisclosed A. Please provide some clarification and we can help you further.

Undisclosed A Manufacturer -- Can you verify if the same statement regarding processing times would apply to Briefcam as well?

Briefcam in particular is one that takes a while to churn through stored video.

I have not really played with the Briefcam product in a while, it's possible the latest version has gotten better in that regard, but I still would not expect anything along the lines of near-instant results.

As a rule of thumb, you're looking at a processing time of about 2x-5x the during of stored video.

Assuming the analytic can run realtime and produce events which trigger recording, isn't the worse case scenario 1x? (By re-streaming it if necessary). Which would then create clips of the detected events.

You are correct, but OP here is talking about stored video that has not already been analyzed/tagged.

Generally speaking, the most effective approach to doing any kind of analytics on stored video is to have tagged the video in real-time as it was recorded.

Let's say you want to do a retroactive search for video where a person got out of a car and approached a bike rack (presumably you have a bike-theft problem on your university campus).

If you have a basic meta-data database of when/where key objects or situations occured you can then do things like query the database to say "give me timestamps where a vehicle was in the scene and a person was in this region (near the bike rack)" (simplistic example). You might get 100 results over a 48 hour historical period, but you'll get those results almost instantly. Next you might use PC-side software to look at those clips and see if the trajectory of the person originated near the vehicle and moved to the bike rack. That will take a little more time, but still not very long because you have pre-tagged meta-data. Ultimately you'd probably get results back on your query in a minute or two.

If the video is not already tagged, then to do a search like this you have to go back and analyze all the video, tag everything, build a database, etc. So, searching 48 hours of video will take at least 48 hours (to process/tag the video), plus some additional time. Maybe it's a little faster if you did motion-only recording in the first place and you try and do some logic to skip parts of the scene that wouldn't be of interest, but you're still talking about query response times measured in multiple hours instead of minutes.

Ok, thanks! That helps me understand where you are coming from.

My reading of the question was that the OP is trying to reduce the amount of data that he needs to keep around, by keeping only the parts that are of interest.

He, IMHO, wants to run an analytic, (for simplicity sake let's say VMD), on the stored video and eliminate the footage that has no motion (or whatever analytic he is interested in).

I believe Agent Vi can do analytics of stored video. I think it was mentioned in a promo for them. I don't know how good they are.

Kind of Video Analytics:

Entering/Exit (People entering or leaving and area)

Traspassing

Motioh detection on certain areas of the video

Nerio, are you saying that you want to keep only the clips of when those analytics are triggered?

So you can reduce the 100TB storage to something less?

Also, what VMS are you using?

Yes

The files are in avi

We want to be able to find events and store the clips.

Lets say that we want to find incidents that happenned 2 years ago and we have the video stored

Our system have a propietary format but we can convert to avi

Note - I work for Aimetis.

If you can play the video file and use a video composite out and connect to a bnc, you can connect the bnc to an encoder which would allow you to send the video to Aimetis Symphony Enterprise Edition which would analyze the video to produce the results you want.

"use a video composite out and connect to a bnc, you can connect the bnc to an encoder which would allow you to send the video to"

I think that's a real bad approach.

  • Quality loss, especially if this is megapixel video, but even analog video is going to be degraded by decoding and re-encoding.
  • With 100TB of video, this is going to take man years to do.

I am not sure what the ideal solution is but the converting to analog out to re-encode and analyze strikes me as the worst case scenario.

NLSS has the ability to run forensic video analytics on video that has not yet been tagged/processed.

They also support the analytics you mention. The big question is how to import the raw video into their system. I have a message to them about this, will let you know what they say.

Re-streaming into a VMS using your .avi files (VLC using RTSP) is possible but it does look like a job that will take several months to run, even processing 50 or more streams at a time.

I can give you a rough estimate of the time and equipment but first you would need to provide for each camera/stream recorded the following:

  1. the elapsed real-time of the stream
  2. the total size of the stream
  3. frame rate of the stream
  4. CODEC info

Also, what is the proprietary format that they are stored in now? All though they might have to be exported to avi regardless, there could be ways to handle them quicker in native format. Like if they are MJPEG for instance, or some standard format, only trivially obfuscated. Also the vendor may have some internal tools or consulting services to aid this.

Thanks