Isn't it about time?

[Detlef Paschke]

Hello everyone,

I'm wondering why a stable 1.37 with the things that work so far hasn't been released, and then work on a new development branch continues. How long has it been since the last version jump?

The reason is quite simple. No one should kid themselves, but Zoneminder's motion detection is unrivaled these days. Any cheap IP camera has better motion detection with human/vehicle/animal selection than you can achieve with Zoneminder and defining zones. Add in a spider web at night or shadows in summer, and you'll have several hundred false recordings the next day. And no, you can't always rule that out by defining zones!

These days, it's best to let the IP camera do the motion detection (it does it much better than Zoneminder anyway), and Zoneminder works as an NVR via ONVIF_Event_Listener. But ONVIF support isn't even available in 1.36 yet, and officially, you're still supposed to use the ancient 1.36 for a productive system. I've been using version 1.37 for years (current version: 1.37.66~20250604.311-bookworm) and only upgrade rarely, and only when no new issues are reported in the forum for a while, and everything runs quite reliably here.

It can't be ruled out that more and more "customers" are being lost to, for example, Home Assistant with its excellent camera integration because ZoneMinder has been treading water for years.

Just food for thought...

Best regards,
Detlef Paschke

[PencilGeek]

I think I came to the same conclusion that the smart/AI motion detection in the cameras is really good, and using ZM zones wasn't working as well. I'm a ZM newbie, so I'm not going to throw ZM under the bus and say it's their fault.

I can't seem to get ONVIF to work for Amcrest 4k cameras. I'm in a discussion with their support (and by proxy their engineering department) to prove to me that it actually works. Not to hijack myself, but if anybody has examples, I'd like to see them.

So instead, I found the Amcrest APIs that allow me to directly query the camera for events. It's virtually zero latency because I have the detection in a separate "producer" thread which pushes the events to a FIFO. Then a "consumer" thread gets instant notification (near zero latency) when an event is pushed to the FIFO, where I then notify ZM with the "TRIGGER" mechanism on TCP port 6802. Inside the cameras, I've got zones and sensitivities set separately. Using the Amcrest API, it will even tell me which zone got the trigger. The camera triggers in ZM tell me which zone, and severity. I've got 10 cameras, and I got this working about 4 days ago. So far, it works amazing. Kind of like ONVIF (as I'm assuming this is how it works), but without ONVIF.

[iconnor]

Why not just release it? Because we know it has issues that we would like to resolve before release. Also, I am overworked and under paid. I'll get to it soon.

ZM motion detection does still have it's uses, but yeah is eclipsed by real AI, and yeah, we SHOULD be pushing the analysis into the camera. Hence why we do support ONVIF, Amcrest, Hikvision and other APIs to do exactly that. But not all cameras are created equal. Hence, we exist.

[PencilGeek]

Why not just release it? Because we know it has issues that we would like to resolve before release. Also, I am overworked and under paid. I'll get to it soon.

ZM motion detection does still have it's uses, but yeah is eclipsed by real AI, and yeah, we SHOULD be pushing the analysis into the camera. Hence why we do support ONVIF, Amcrest, Hikvision and other APIs to do exactly that. But not all cameras are created equal. Hence, we exist.

Are you saying there's Amcrest API support directly built in to ZM? If that's what you're saying, I'd like to try it to see how it compares with my home-grown approach described in my previous message.

[mikb]

we SHOULD be pushing the analysis into the camera
...
But not all cameras are created equal. Hence, we exist.

It's great that "AI" cameras exist, assuming they are self contained, and not dependent on some questionable security/availability cloud-crap, subscription-based offsite AI for this detection.

But for people that are using cameras that are just that -- cameras -- ZM ditching doing motion detection would be an instant fail.

[Detlef Paschke]

Why not just release it? Because we know it has issues that we would like to resolve before release...

Hello,

My concern is that Zoneminder will lose more and more users if we continue to stand still.

For example, it's becoming increasingly difficult to integrate a modern IP camera into Zoneminder so that it runs synchronously. With the old 640x480 or, for that matter, 720p cameras, this wasn't a problem. But with current 2K, 4K, and who knows how many more "K", the camera's FPS has to be continually reduced so that it doesn't have a time lag of several minutes in Zoneminder.

It shouldn't be the camera that needs to be adapted to Zoneminder, but rather Zoneminder that needs to be adapted to the current technology. Such things are much more important for retaining your users than any optical gimmicks.

Best regards,
Detlef Paschke

[iconnor]

Who is standing still? Not many people are working on ZM anymore. People come and go. I remain but it is more than I can handle.

Who are you? Are you offering to take on some task to help out?

I commited to releasing 1.38 in May... came and went. It is very close but I have to do things that pay my bills first and foremost, so the remaining tasks have not gotten done.

[PencilGeek]

Why not just release it? Because we know it has issues that we would like to resolve before release...

For example, it's becoming increasingly difficult to integrate a modern IP camera into Zoneminder so that it runs synchronously. With the old 640x480 or, for that matter, 720p cameras, this wasn't a problem. But with current 2K, 4K, and who knows how many more "K", the camera's FPS has to be continually reduced so that it doesn't have a time lag of several minutes in Zoneminder.

I was curious about this as well and I found another ZM user come up with -- what appears to be -- a very novel solution. Keep the camera settings at full resolution (e.g. 4k). But the ZM settings should be much lower, maybe 15%, or 25% of that. In my case, that meant setting my ZM resolution to 720x405. The analysis is performed at 720x405 allowing the frame rate to keep up; but when an event occurs, the video is saved at full resolution. I tried this approach, and it worked fantastic.

It shouldn't be the camera that needs to be adapted to Zoneminder, but rather Zoneminder that needs to be adapted to the current technology. Such things are much more important for retaining your users than any optical gimmicks.

Best regards,
Detlef Paschke

That's just not realistic. If you have an 80386, you can't expect ZM to compensate for your 40 year-old CPU. I have a 4-core (8 with multi-threading) i7 at 3.2 Ghz. I don't expect ZM to keep up with everything I can throw at it. Without CUDA offload, I could max out the CPUs with just 5 cameras. So I tried CUDA offload, and that was very helpful with 10 cameras, but not perfect, and I could still hit an occasional stall. I still had to play a few tricks to allow my 10 year-old i7 to keep up. So, I experimented with my own approach to use the camera APIs to monitor for events and then notify ZM that the event has occurred. That's working perfectly and my CPU utilization is now down to 13-15%. Problem solved.

[iconnor]

There is lots of work being done to make this all better. 1.39 should reduce our cpu/ram use A LOT. Also, support for hwaccell which MOST desktops have in one way or another.

We are also going to have built in support for AI chips that no one else has.

So it is going to be fun... but I need to do the work and get it done.

Also, live viewing with jpeg streams just needs to end. So with some browsers now getting h265 playback and av1 becoming mainstream, webrtc is becoming viable and we are going to support it.

The OP is right though, need to get 1.38 out.

[xobes]

All this talk about letting the cameras do the work is rather concerning to me. Sure, fine, whatever, let it happen. But I, for one, have isolated my cameras to their own lan and firewalled that lan from being able to get out to the outside world.

That said, perhaps the cameras really are doing their own processing and not relying on cloud serviced (but I doubt it).

Looking forward to 1.38/1.39 and wondering how I might be able to test those out. I would really like better object detection, and frankly, better ability to debug/diagnose and configure the alarm settings.

[Detlef Paschke]

That said, perhaps the cameras really are doing their own processing and not relying on cloud serviced (but I doubt it).

For example, a Reolink E1 Outdoor SE PoE
Almost perfect detection with PTZ object tracking, separate settings for people, animals, vehicles, or general motion, and completely cloud-free and cut off from the internet.
I'm currently only using it directly with Home Assistant.

My Annke C500 also has quite good detection, separated by people, vehicles, and general motion, and completely cloud-free. They're working with ZoneMinder.