What Video Analytic Can Detect Violence, Swearing, Shouting, Hitting, And Running?

We would like to detect Loitering, Group Formation, Agressive Behavior, and violence. Markers we would be interested in knowing about are swearing, shouting, hitting, running etc.

All insight would be very much appreciated.


Colin, how about something easier like seeing through a brick wall? :)

There's nothing that is fully proven / bulletproof that does this. Whatever you look at, please thoroughly test.

The key elements will be: (1) what tolerance for nuisance / wrong alerts will they have? i.e., if the guard gets 20 incorrect alerts a day, can they handle it? (2) how much money are they willing to spend? this is going to be a premium $1,000+ analytic software type solution (or more).

There are some companies that claim to detect abnormal behavior, like BRS Labs and iCetena. This is not specifically violence or aggressive behavior, but presumably such activities would be considered abnormal.

Hopefully members have suggestions. If you do have suggestions, please provide some details on how they might fit.

Thanks for the response, I do have some tolerance for over alerting but less tolerance for missing the incident. We will generally budget about $5000 for four streams independent of the NVR or camera and then about $1000 per stream after the first 4.

Colin, if you don't have much tolerance for missing incidents, you'll likely get a lot of nuisance alerts (i.e., people just hanging out or having a catch gets mistaken for aggressive behavior, etc.).

I think your budget might work but it's probably on the low end money wise for this type of application.

Let's say I can increase my budget by 50%. What would be your recommendations.

I'd start the two mentioned above but the problem is BRS Labs has all sorts of yellow, if not red flags, as a company. iCetena seems to be a more reasonable / conservative company but they are fairly early stage as well.

The challenge here is that the 'leading' video analytic companies - like VideoIQ (now Avigilon), ioimage / DVTel, Agent Vi, Sightlogix, etc., etc., don't even claim to do this.

Hopefully someone has some niche recommendations of a company who is focusing on this. Expect it to be server based, requiring on site tuning, etc.

It's very difficult to tell if someone goes beyond the limits, even to human beings. It is a very subjective thing to tell..

Colin is audio available or only video? If audio is present there are a couple companies at least claiming aggression detection and the like. This video is from Sound Intelligence:

This one is a BBC video about a company called AudioAnalytics. No idea how well they work...

Are you expecting Ray Rice at your facility?

Axis technically has audio detection in some of their cameras that you can set up to a certain threshold before you trigger an alarm. However it wouldnt obviously detect the difference between talking really loud/happy yelling vs swearing and shouting... if you find that product, let me know because every school ever built needs that product.

Btw, related, there is an audio analytic available as an application for Axis cameras. It is Sound Intelligence's Advanced Audio Analytics which claims:

"Advanced Audio Analysis can identify specific sounds, such as aggression (stressed voice), breaking glass etc. out of a mix of ambient noises. Even at a lower volume."

How well it works, I have no idea.

Dear Colin,

Sound Intelligence has installed thousands of sensors around the world for aggression detection (and other audio analytics). Like video analytics, audio analytics is no magic, however we do get very good results detecting aggression with our application that runs on certain Axis cameras.

Please contact me at derek.vandervorst@soundintel.com if you would like to discuss details.

Kind regards,

Derek

I've a customer using iCetana a lot, he claims that it's not useful for schools for example, but works pretty good in malls (big retail stores). I can provide an intro if you want.

I know that www.audioanalytic.com works pretty good as well for shooting etc.

We've some apps in our app-store at www.angelcam.com that can many things, e.g. counting people on the scene (-> group formation?), measure the speed of the object (-> agresive behaivour?) and time spend by particular object on the scene (-> loitering). I would like no know more about your specific request to provide more accurate answer.

Hi Colin,

My company Bootpin builds video analysis software that may address some of your requirements. The software can detect activities in video with fairly high specificity. For example, the software can detect and differentiate between running, kicking, punching etc. The software uses models built with machine learning and can be trained to detect practically any activity of interest. Swearing and shouting however are a bit too subtle for reliable detection through video alone.

If you are interested in a demo or more information, please contact me at adeo@bootpin.com .

Ade, thanks for sharing.

Here is a video I found on your site:

What resolution / scene width / angle of view / etc. do you need to deliver this accurately?

(I'm the Customer that

We use a product called iCetana that covers almost all of your requirements. A direct clip from our website is below. It trully is an amazing product.

iCetana { eye-set-ahna } is derived from the Sanskrit word “Cetana” meaning “visible”.

iCetana is a software company based in Perth, WA that has developed a video analytics application based on a patented innovative technology for “anomaly detection in large data-sets” developed and maintained by Professor Svetha Venkatesh and her research team at Deakin University. The software application automatically learns “movement patterns” in a scene and then analyses live feeds from multiple video cameras to detect and send alerts of any “abnormal movement behaviour” in real-time.

For more information about iCetana, please feel free to visit the iCetana website.

Why did Campus Watch select iCetana?

Providing full time “human monitoring” of campus-based network cameras is beyond the means (and requirements) of most schools. Even if full time monitoring staff were available, the sensitivity associated with the ongoing surveillance of school children is difficult to ignore. Through sophisticated learning algorithms, the iCetana software suite possesses the ability to simultaneously scan the video stream of a large number of video cameras and selectively display any “abnormal movement behaviour” occurring within the campus community. iCetana’s ability to only “focus on the one percent” of abnormal activity occurring at any one time effectively removes the stigma surrounding the continuous surveillance of children while providing the school with yet another valuable tool to enhance their duty of care capabilitites.

Colin, us (TechnoAware) we can actually offer video analysis functions for detecting those events: functions Loitering, OccupancyRate, PanicDisorder in particoular.

Nevertheless, I fully agree with John about the complexity of those behaviour analysis, above all in crowded areas. All depends on the complexity of the scenario. To have for example an efficient loitering detection (read long term tracking of a subject..) among a crowded area is, at today, still too challenging to have useful results.. (Even if I remember to have seen more than "someone" claiming its fully compliance in some absurd tenders (above all from Middle East) requiring multi-camera tracking among crowd.....;))..)..

On the contrary, for example, a group formation might be anyway easier to detect. While a panic/aggressive behaviour could be easier to detect in a place where people may be even a lot, but reasonably behaving with a regular dynamic: inside a stadium, for example, it may have no sense; but outside the stadium maybe yes..

As John wrote, "Whatever you look at, please thoroughly test.".

Thus, more than just claiming, as we usually do what we can offer to you is a free demo and all our technical support to test it directly in "your" scenario. Or, do you have videos from that scenario that we can try to process together offline to see what happens? Then, let's see in real life if we can achieve useful results or not.. Please, feel free to contact me for that.

About the price; well, I think we definitely meet your budget....;)

About Audio Analysis, we can't provide that but there are technology partners doing that with which we could integrate. It surely could be very interesting to integrate video and audio analysis, also right to attempt to reduce the alerts. For example, you could have a middleware correlating video and audio analytics (perhaps any "open" VMS can already do that..) and define a rule such as "aggression alert + shouting within 10 seconds = ALERT".

Merry Xmas to you all guys, have a great time

Simone

TechnoAware