Person Detection Update: A New Experiment for Premium Features

Sounds like a political association… Appropriately named too


And even if it correctly detects a person the soonest you will know sbout it is ~13 seconds, but more likely 30+ seconds, after the fact.


Their pay structure always reminds me of…


I am confused… Does anyone think this is acceptable???


Recently, I realized how bad the delay was. I set up a lamp in my home office to go on for 5 seconds when a camera facing my front door detects a person. On average the lamp turns on 15-20 seconds before the notification hits my phone. So it seems to be all about the push notification, whatever processing is going on in the cloud to handle triggers is happening in a timely manner.

Wyze in general is slower on the server side then other smart devices, even when working well.

I used to have a Wyze bulb in a lamp and a Wyze plug controlling another light with a plain lightbulb in the same room. Both were triggered by an Alexa routine when a Zigbee motion sensor was triggered upon entering a room, but it took 3-5 seconds when things were working perfectly and I could be half way across the room when the lights came on. When I switched out the Wyze plug on one light to another brand, and left the Wyze bulb in the other lamp, the light with the other brand would come on nearly instantly when entering the room and the Wyze bulb always lagged a few seconds behind at best.

After seeing that slower response time, I also eliminated the Wyze bulb so no Wyze products are in that loop at all and both lights come on within one or two steps entering the room.

I do not know if their protocol and process allows (another example a closed systems deficit as we as users are blind as to cause let alone fixes) , but it’s strikes me as addressable by having the process hand off a trigger which would generate an alert by the subsystem monitoring for events at various points during the process instead of sequentially after generating the uploaded video.

A motion event is going to be processed, it’s final disposition as a Person is just a classification of the event. Handoff to an edge processed monitor loses significant time if all verification occurs after upload of all data to the cloud.

While doing analysis of presence alerts and trying to identify options for my system, I looked at a few aspects impacting my results.

On my local system, I run a Pi-Cam on a RaspberryPi zero (I also did testing with RTSP feed from a Wyze cam, but abandoned it as I wanted to stay with native Wyze and the rtsp was a beta product). The local system events are processed by a tensor based analysis MotionEye living on a Home Assistant controller, video data is archived to a usb ssd attached to the same Pi ( if you want a larger archive I’d bring up an NVR, but I’m okay with a couple weeks rolling capture ).

The local alert system simply waits for an MQTT message then sends out a push notification, then sends a TTS alert to my Alexa ( I could achieve a faster response using Almond voice assistant as it’s local, but I have many Alexa devices around the home, so I stuck with Alexa) , then my Telegram server bundles the video clip and sends out its actionable alert.

My initial tests proved that my setup of 8 cams that are being monitored is well within the scope of this setup. The accuracy, while not perfect, is better than any results I have with Wyze.

Using this method, you can achieve great results, If you are willing to put in the time to do a bit of reading and deal with what I would certainly categorize as a living system as Home Assistant will update and patch as much as your willing. This is fine for individual instances, but to try to replicate at the consumer mass market level is a problem waiting for a solution. I believe that the HA ecosystem will reach various points where a stable demarcation can be embedded in an appliance server, but I don’t see that point as attained yet.

How could Wyze apply the lessons learned from XnorAI or the many far more knowledgeable users of HA and other Edge systems?

Personally, I would like to see someone like Wyze be more than a bundler, more than simply maximizing sales of a few discrete SKU’s.

Wyze’s mission statement of bringing Smart Technology, regardless of classification or narrowly defined type to a broader market, being able to provide a pricing model that fills a segment not serviced by others in the industry has merit.

I am not convinced that closed ecosystem will ever be an approach that dominates a market that has existing standards and dominant players with the advantage of capitalization and the ability to simply acquire threats if they encroach on marketshare.

I know that the average consumer of smart devices wants to simply plug them in, and with as minimal setup as required, enjoy the conveniences they bring. A $20 smart video camera is not expected to outperform a security grade hardwired surveillance system, and it would be foolish to try as this is not expected by a massive consumer base that they wish to service, but basic experience with inaccurate motion alerts will quickly remove any perceived convenience the user expected.

I think that there is a hole in the market yet to be filled, If Wyze would develop a hub, one that fills the established request for local storage of videos from the local cams, populate with a SBC controller such as RaspberryPi, partner and then bundle in a Coral TPU, it would provide the integration that would differentiate them from anyone else currently available.

Just a few thoughts

If you could point to one aspect, what makes WyzeCam slower than others?

I would guess it’s on the Wyze server end. I don’t know all the details or sequence of events that happens between the time a Wyze device triggers and the notification happens or a rule or routine executes. But I think it has to go to the Amazon or Wyze servers and then back to the app. I don’t think notifications or actions happen locally on your home network, I think it has to go out on the net and back.

No, person detection is only on cloud now, I’ve not had a chance to see if the PD is slower than, for example, another Cloud based system, but edge process has an advantage that is sure, I don’t use my PD for time sensitive announcement or control, I use PIR for motion lighting since I really don’t care if it’s an animal or person, I have the lights kick on, even without the PD aspects, I see a big difference in response over other cloud base systems, but don’t know what the root issue is,

The cloud based system I was actually really amazed by, even though I flashed the firmware for local control, is Tuya Smartlife products. They are a no hub cloud system, and their response to control signal was significantly faster than Wyze, I have one non-converted unit I keep for testing, and click to light on is less than 1/3 second consistently.

I really don’t want Chinese servers with presence in my house (lights only no cams) so I flash them with Tasmota for local home assistant control, but they are hands down the fastest cloud only system I have tried.

I have several smart devices on the Smart Life app. Some are Tuya, and others are branded with other names like some of the Amazon special deals of 4 smart plugs for $22 or something like that on their Prime Day deals. They work really well.

Yes, I do know that person detection is cloud only. What I meant was even Wyze Sense contact/motion sensors and maybe bulbs and plugs need to go out to Wyze servers before sending a notification or turning on/off from a Wyze app Rule or Alexa Routine. That would explain why my routine triggered lights stopped working along with notifications during those times Wyze said the problem was Amazon server maintenance.

I believe in Wyze products (think hardware)… I don’t have the same faith in their firmware or support systems (think Servers)

No, I get that, I guess my point was that there are cloud based systems that are much faster than Wyze, Tuya among them, so that means that Wyze has the potential to be much faster even though they are cloud only.

Not meaning that Wyze is fine, meaning that the problem is not the cloud based processing due to the hop time, that the issue is bigger not just due to a 200ms hop time.

The Tuya servers are outperforming the Wyze and last I checked, I don’t think there process servers are US based, so they are outperforming despite the faster hops that Wyze has. I have spent some time, but its difficult to disgnose Wyze issues as we have no way of seeing any event transactional data. At least when I’m diagnosing say SmartThings, there is a log can look at, and if I’m using WebCoRE, I can get line by line logging, memory, variable content, etc.
Wyze has a closed system, but more they do not provide access to logging, there is no way to even see the call time verses the response time, so you at really shooting from the hip when trying to analyze any process controlled by Wyze. This is one of the key reasons that I do not use Wyze rules, even to control Wyze to Wyze devices (wyze sense contact or motion sensors controlling Wyze bulbs, I will control the logic from Alexa routines so I have visibility of the transaction. They are a bit better now, but when your dealing in a complex system, the fact that the rule executed is not enough, you need timing and control state logging which Wyze doesn’t have open)
I use Home Assistant for a majority of my controls. It logs and can log at defined levels down to the discrete level.

One of my concerns about Wyze trying to become a complete Automation platform, especially after they announced the Alarm system which means now they will have users, misunderstanding what Wyze is offering and perhaps putting their safety in the hands of an Alarm system that doesn’t have any way for their users to verify logs to detect issues before they impact their safety…
Yes, Wyze has said that in regards to their cams that they are NOT a security monitor system, they are a Smart Camera system that gives users capabilities they wouldn’t necessarily have access to otherwise…
And then they announced an Alarm system that those same users will rely on for safety. Additionally, although its a thin line, the fact that they stated that they were not a security company and they were not selling safety monitoring systems, was enough to hang a legal shield on…the users that use the cams in a security application now, if issues happen, I’m concerned that Wyze just blooded the waters with chum and a far easier case can be made.

Sorry, rambling, bottom line, there is a lot of improvement needed to be a timely process control, let alone a security system defending people’s life and or property.


Yep, THIS is what the bottom line of my extremely long and rambling answer.!!

Although the volume of information in your post contains lots of information And factually it makes sense but your 2nd to last paragraph truly addressed the problem we, as users face…
Boiled down to more palatable phraseology, we want dependability and we want truth in advertising. Period

1 Like

Concise wordsmithing, NICE,

Regarding chinese presence, I’m sure it’s already there in many unsuspected areas

Well, perhaps, but somehow giving them direct access to control items in the house seemed a bridge worth busting. Yes, reality is probably far more concerning than that…but we do what we can…

After this thread, I am once again wondering why Wyze is in the 6-7 second trigger point to action execution, SmartThings is down to 2-3, yet Tuya is still a magnitude faster.
They must have a well resourced server system and far better coding and processing (think lots of video being processed in their control zones.)

Kind of a disturbing question. Wyze is either part underspending and not specifying servers capable of handling the loads in a functional time, this may be purposeful due to their misunderstanding on costs for cloud based control and are purposely under performing to lower this expense…,

George Orwell wasn’t that far off…