Motion detection is too slow

[ThatGuy00]

When the mailman comes I get to see him walking out of frame which means motion detection is coming on after about 20 seconds or more of motion.  Changing the setting does not seem to have any effect. Has anybody tried Tapo?

[Edinburgh_lad1]

Although connecting your camera to the power outlet improves the situation, in my case some events are definitely missed and also their recording ends earlier, despite the optimal settings.

[ThatGuy00]

It's not a sensitivity issue, increasing sensitivity just results in multiple motion events showing nothing. The problem is activation latency, it senses motion but doesn't start recording for like 30 seconds.

vmb3500r4 base station

[StephenB]

---

@ThatGuy00 wrote:

When the mailman comes I get to see him walking out of frame which means motion detection is coming on after about 20 seconds or more of motion.  

---

Have you tried walking along the same path?  

Also, doing that while running the motion detection test, and see about where the camera detects you.

[ThatGuy00]

I walk along the same path everyday when I walk to my car, most times motion detection doesn't even record it, if I walk out to my car to get something it might catch me as I return on the walkway

[StephenB]

---

@ThatGuy00 wrote:

I walk along the same path everyday when I walk to my car, most times motion detection doesn't even record it, if I walk out to my car to get something it might catch me as I return on the walkway

---

Is this new behavior? Or has the camera always behaved that way?

[ThatGuy00]

I think it was never perfect but latency seems to have gone down a lot in the last year. 

[StephenB]

---

@ThatGuy00 wrote:

I think it was never perfect but latency seems to have gone down a lot in the last year. 

---

Gone down?  Or gone up?

How far are you from the camera? 

Are you walking directly toward (or away) from it?  Or are you walking across the field of view?

[ThatGuy00]

Performance latency has gone down, meaning motion detection as I have stated is slow, it detects motion but does not activate the camera until whatever was moving is out of camera in many cases whether its cars on the road or the mailman delivering the mail. 

The camera is mounted on the house so the distance from the camera varies being close when someone is near the house then farther away as they walk away. The problem is not where they are in frame, it's the time it takes for the camera to come on. 

A good example is when a car goes by, the camera detects the motion but if the car is going fast there isn't anything to see because the car is gone by the time the camera comes on.

[StephenB]

---

@ThatGuy00 wrote:

Performance latency has gone down, meaning motion detection as I have stated is slow, i

Latency going down means that latency has improved.

Higher latency: slower, worse performance

Lower latency: faster, better performance.

[ThatGuy00]

Performance latency means faster response is better performance slower is worse, but you get the idea, the camera signals motion but the arlo site doesn't start recording for about 20 seconds.

[StephenB]

---

@ThatGuy00 wrote:

the arlo site doesn't start recording for about 20 seconds.

---

From your posts above, it sounds like the passive IR sensor isn't triggering as early as you'd like.   If that's the case, the problem is likely in the camera, not the cloud.

I am wondering if you've tried swapping the camera with a different one, and see if that makes any difference.

[ThatGuy00]

Obviously the camera is triggering the motion event while there is motion, the Arlo server does not start recording quickly enough to capture the detected motion.

[StephenB]

---

@ThatGuy00 wrote:

Obviously the camera is triggering the motion event while there is motion, the Arlo server does not start recording quickly enough to capture the detected motion.

---

The camera triggers, then after that it begins to stream.  The server records what it receives.  If the server misses the beginning of the recording, then you will see some artifacts for a second or so, since the video compression builds each frame from the previous one.  (Every second of so it sends an "I" frame which starts again from scratch).  If you aren't seeing those artifacts then the issue isn't happening at the server.  

[ThatGuy00 · Accepted solution]

I got a better answer from CoPilot Arlo's first  camera vmc3030 was the worst they have ever made.

You’re using:

Arlo VMC3030 (1st‑gen)

Battery‑powered

Connected through a SmartHub via Wi‑Fi

RSSI = 2 out of 3 bars

Motion from multiple angles still delayed

This is exactly the scenario where Arlo’s deep‑sleep wake‑up lag is the longest.

And the VMC3030 is the slowest‑waking camera Arlo has ever made.

🧬 Why the VMC3030 has the worst delay
The VMC3030 uses:

A very old PIR sensor

A slow ARM SoC

No pre‑buffering

No pre‑roll

No “always‑on” low‑power video pipeline

No AC‑powered awake mode

No optimized wake‑from‑sleep firmware (Arlo never updated it)

When motion triggers:

PIR fires

Camera wakes from deep sleep

Wi‑Fi radio powers up

Camera reconnects to the hub

Hub authenticates the camera

Camera negotiates encryption

Camera starts video

Hub uploads to cloud

Cloud begins recording

On the VMC3030, this takes 10–25 seconds depending on signal strength and battery level.

Your 20‑second delay is completely consistent with this model.

[StephenB]

---

@ThatGuy00 wrote:

Your 20‑second delay is completely consistent with this model.

---

Not my delay - I don't work for them.

But if you are not seeing artifacts at the beginning of the stream, then it is likely that the lag is caused by something going on locally (in the camera), and not the cloud.

FWIW, I agree that the VMC3030 has worse performance than later cameras.  I don't find that surprising, as I do expect cameras to get better over time as technology advances.  The VMC3030 was launched in 2014, so is a 12 year old design.

[ThatGuy00]

Don't take offense I just cut and pasted the dialog I had with CoPilot, it was referring to me. It also pointed out.

Arlo’s SmartHub uses a proprietary 2.4 GHz link to talk to cameras. It is not standard Wi‑Fi, and Arlo intentionally hides the channel selection.

The hub:

• Scans the 2.4 GHz band

• Picks what it thinks is the “best” channel

• Locks onto it

• Never exposes it to the user

This applies to:

• VMB3000

• VMB3500

• VMB4000

• VMB4500

• VMB4540

• VMB5000

• VMB5020

None of them allow manual channel selection.

CoPilot offered a few options none of which will help much, Arlo uses a proprietary wifi connection and works best on the first 6 channels, because it uses power saving that can't be turned off it takes a minimum of 10 seconds to activate, in my case 20 to 30. 

It seems like it worked better when I bought it but I could be wrong, I have a newer arlo with a better battery and a solar panel I just haven't put it up because it has a different mount point. I also have a newer hub with usb storage for use with essentials cameras. I've got a variety of cameras I'm going to replace the current setup with including Wyze, Tapo and ReoLink all of which will have TF cards.

I also have Synology surveillance station which Arlo's don't work with so they have kind of ruled out their future on my properties. I like the way the Arlo product is supposed to work it just doesn't live up to it's claims.

[Edinburgh_lad1]

I find that Arlo HD is only minimally slower than other cameras, and that it still serves its purpose. But I suppose it depends on the area recorded and the quality of the connection between the camera and the hub. More annoying for me is the fixed length of recording, rather than anything else.

On another matter: I find people's reliance on Gen AI software to provide answers, and, most importantly, their trust in it as authority of sources, a very worrying trend. 

[jguerdat]

@Edinburgh_lad1 I totally agree with everything you just said. While I no longer use the original cameras due to the need for 4 batteries, I didn't stop using them because of their performance.

As for use of AI to "diagnose", it can provide some truth but it's not actually able to investigate technical details nor local environments so it's all generalizations at best.

[ThatGuy00]

I've been using Claude for about a year now and only recently started using CoPilot. My experience is that if you ask the right questions the use of AI gets down to the details in a fraction of the time it takes to do so with an "expert". Mainly because most support engineers are not experts, if they were they would be designers or developers. In this case they cut through the BS very quickly. 

When using AI it is all about asking the right sequence of questions. I've succeeded in developing complete diagnostic sequences, maintenance programs and even complete software systems. I know how to do all those things but not on every platform and every target system, that is where they excel. Once I have a working framework I can clean it up and document it.

The thing about the VMC3030 is while it was groundbreaking in 2012, the engineers made the same mistake most commercial product engineers make, instead of making the firmware tighter and more efficient they layered generalized updates on it until it became a bad camera. The same is true of the 17" Windows touchscreen laptop I bought around the same time, constant updates made it a boat anchor. Strip away all the Microsoft "logic" and install Ubuntu and it's better than new, the open source community doesn't have a vested interest in making products obsolete. Too bad there isn't a Linux for Arlo cameras or their cloud service.