Asking for a friend..
Regarding event detection on a typical camera install:
With advanced detection features such as Line Crossing, Intrusion Detection, Face Detection becoming norm, do you really ever use these features on a frequent basis? The key word being frequent.
Or, do you just still use good ol fashioned motion detection in which you block out areas of the image in which you dont want motion to occur and adjust sensitivity as needed?
[IPVM Note: Video Analytics Usage Statistics 2018]
The short answer is no. It is not something we frequently use. However, if the scene is friendly to line crossing analytics we have had success utilizing it on 40-50 cameras. The most success is well maintained fence line on thermal cameras. Axis VMD 4 is one we use more regularly though as it is free and better than nothing.
*Note I work for large end user not integrator.
I think part of this question is why not. There are a lot of blocks to using analytics in various applications, like events not passing to mobile apps while simple VMD does, VMSes not supporting third party analytics, or simply lack of marketing of these features.
I think if they were seen as more of a "default" than VMD, they'd get used more, but there are issues blocking it. For example, Hikvision intrusion analytics aren't extremely accurate in comparison to some, but they are still significantly more accurate than VMD running on the same camera. It would make sense to me for a Hikvision camera connected to a Hikvision NVR to default to the analytic first, since it's more accurate, then fall back to VMD if desired.
But that's just me! Anyway I tended/tend to use them whenever possible, but I'm a nerd and really into optimizing cameras I touch.
The challenge for us has not been integration with VMS as we are pretty lucky in that aspect. It has just been reliability of performance. We operate in all kinds of areas and in some it works great like the previously mentioned fences with well maintained landscaping and in industrial settings where shadows and shrubbery is minimal.
However, at normal office locations or on the outskirts of these industrial areas it is very challenging due to shadows of trees, animals in the area, etc. all things that are not handled well by the cheap cross line detection that we used at the time. To compound this adjustments must be made seasonally as the position of the sun changes to avoid false alarms. There are arguably better solutions now but after years of abuse by false alarm our officers and leadership just don't see the value in putting out the money required for the truly "good" analytics nor are they historically willing to make the implementation of analytics a talking point from the start.
Knowing that it is a desire and planning for the implementation goes a long way toward having a solid deployment. Slapping it on later is not the best bet.
I guess we are different 90%+ of the cameras we install have analytics and are used in either a proactive or reactive application.
Well every Avigilon H4A camera comes with analytics turned on for recording, searching and appearance search. Most of our customers use analytics for searching. Say you have a parking lot full of cars and you're looking for a person. They will use classified object search to only search for people. Appearance search is also very popular which you can select a person and search for that person across all your analytics cameras. We have some customers using rules to trigger alarms when people are detected to pop up alarm events. We have used the line crossing rule to filter cars entering vs exiting when it comes to searching.
I had one customer with a problem with employees smoking which they wanted to install a camera. We wound up recommending using a loitering rule to trigger an alarm to filter out people leaving the building and not smoking. This way they could stop them in the act instead of just watching them smoke on video. Lots of options when you dive into it.
Most integrators I talk to just use regular VMD do to results, but it's also what they know. VMD sucks pretty much across the board, but setting up camera based intrusion detection opens up a whole can of worms depending on the hardware and then does your software understand that event data.
I will say though, Mobotix has done an awesome job with their proprietary VMD, Activity Sensor, when compared to other camera VMD. I was a skeptic until I started to really use it against typically VMD and it is night and day better.
To me it has to be a mindset change. The techs out in the field barely have enough time to get the system up and running let alone explore the new features such as analytics. Therefore the technicians know how to setup motion so that is what they do. Some of what I do is get these features working. Problem is I wear many hats and can't spend near as much time as I would like, especially showing our technicians what can be done.
Yes I completely agree with this. Weekly training for your techs are a must to keep them up to speed as software and features are updated.
We are starting to do more training internally. Training is one of things that are easy to do and even easier not to do. Problem is we get really busy and we don’t take the time necessary or at least we think we don’t have time. Training for whatever is really important. Even for simple topics. Getting your technicians in for regular training shows them you want them to succeed. One way we are going to make training more engaging is have a technician train everyone else in a topic they know well are are good at.
I use it sometimes. I try to encourage clients to record continuously. For a storage unit we did recently i used line crossing on all aisles and line crossing in the parking lot entrance. It speeds up investigative playback.
In my previous job I had one customer that wanted a bell to ring in their office any time someone entered the lobby, but they did not want any notice when people left. We used an Axis camera with cross line detection only in the direction of entry tied to a relay output that sounded a chime. We did that maybe 10 times, but the use case was specific and the area was fairly controlled so it made the reliability of the detection something I could be confident in.
I also had one customer with Mobotix cameras that LOVED to see the hotspot map of their convenience stores. I managed to sell that to a few smaller local museums too, it helped them to balance their exhibits and traffic flow.
In general though, there are too many variables to make most of these "toys" usable.
Ask questions and get answers to your physical security questions from IPVM team members and fellow subscribers.
