Camect "Worlds Smartest Camera Hub" Tested

That's fair .... We were pretty surprised by those two as well and want to fix them. We've already asked Rob for some more info about those cases. It was especially hard to believe we missed the guy at the door.

The software automatically temporarily suppresses notifications for alerts that are occurring "too frequently" to be useful to an average person, and we're wondering if this is what happened to the guy standing in front of the door. (It would depend on what was happening before this guy showed up at the door.) Suppressed alerts are still reviewable in the UI and if you then rate one such alert as important, it will learn that it's not supposed to suppress notifications, e.g. for people, and it will send them to you no matter how frequently they're occurring.

For the running guy case we've requested the video clip if they still have it. From the image in the report it might have been due to the fuzziness of the image, but this also could have been due to the frequent alert suppression if it occurred after several other tests that detected people in that room.

Hi Ross, I think we already have at least 90% of what you want. See what you think:

1. Both the email alerts and telegram alerts link back to the UI to see the actual video.

2. We have region of interest that you can draw out, although it's hidden away as we consider it an advanced feature from the perspective of the ordinary home user.

3. We have automatic camera discovery for a large variety of cameras, including specific drivers for several brands, and fallback to Onvif and autoprobing of several common RTSP urls when Onvif is not present. Note that IPVM said that RTSP urls may be required for commercial cameras. We haven't tested many expensive cameras an average end-user won't buy themselves, and this is easily remedied. (Adding those drivers is usually much easier than drivers for consumer cameras.)

4. It's true we don't have a native app, but I want to make sure you understand what a Progressive Web App is. It uses the browser to present the UI, but after installation the app is fully standalone like any other app -- i.e., it launches from an icon on your home screen, opens with a splash screen, is a standalone app distinct from your browser in the app switcher, does not share credentials with your browser, and you don't see any browser controls anywhere. (i.e. It behaves a lot like a native app, but subject to some caveats as mentioned below.)

Here are more details complete with all gory caveats:

Email alerts: You get back to the UI by clicking the animated gif... Check out some example email alerts in our alerts demo. You can try clicking the link although you won't have permisson to open the UI. In Android, the link will open the Progressive Web App. In iOS, it opens Safari.

Telegram alerts: You get back to the UI by clicking the list of detected objects, e.g. in one alert in the report above, you see "Wyze person detection cam saw a person" and clicking on "a person" will take you to the video. Telegram currently enforces that the link opens in a browser, but we're going to file a bug asking if they'll allow the PWA to be used when there is one.

Our PWA also has native text-only alerts on Android, but not on iOS as iOS does not support them for PWAs yet. We recommend telegram alerts across the board because the video clip we can send in them makes them much nicer than what the size limits on native alerts would allow for, and 90% of the time eliminates the need to go look at the video.

Region of interest: As IPVM described, the software attempts to automatically learn the region of interest based on feedback you give on the alerts (i.e. click on "thumbs down" and tell it that the alert is unwanted because of the area of the image it came from). However, we do also allow you to draw out one or more regions of interest manually if you want, and even to do somewhat complex logic using overlapping regions with and/or behavior if it comes to that. This feature is hidden away in the UI so it's understandable that IPVM didn't find it. We don't expect most users to use it, but it's mentioned in the docs.

Automatic Camera Discovery: We have several automatic discovery mechanisms. Our specific drivers include Axis, Hikvision, Dahua, Reolink, Foscam, Nest, Arlo, Wyze, Doorbird, Trivision, Ubiquiti, consumer models of Samsung, and some models of Vstarcam, Dlink, and Trendnet. We have a fallback Onvif driver that works for many brands but currently cannot find the media service on some of them. We also auto-probe cameras with open RTSP ports for several common urls, .e.g although we're never recommend them, the large number of XiongMai-based cameras out there will work.

Since our drivers only support the basic stuff (get the video stream, move the PTZ, and sometimes set/get the camera name and set wifi credentials) it's very easy to add new drivers. We haven't tried several of the commercial cameras that IPVM used (as no one in our target market has them) but it's much easier to add drivers for these cameras than for the average consumer camera. The large brands are recognizable from MAC address alone, and usually have well documented RTSP urls, so we can quickly add the right RTSP urls for any big brands.

IPVM has not shared details of the camera problems they ran into with non Hikvision/Dahua, but I'd guess that they had problems with the Onvif driver above and with cameras that don't use the list of common consumer RTSP urls we probe. It's also possible that our discovery had some problems due to the number of cameras on IPVM's network, which was much larger than you'd find in a typical home or small business.

Native app vs Progressive Web App: As I mentioned above, the PWA is actually quite a lot like a native app although you install it differently. The biggest caveat is that iOS does not support notifications for PWAs -- but Telegram produces much nicer alerts anyway.

What you see in the app looks like the mobile web UI minus all the browser chrome. That's something that many people like, because everything works the same way as it does from your desktop/laptop. We're aware of several shortcomings in our UI -- but those are our fault, not the fault of it being a PWA versus a native app.

Are there specific things you worry about the PWA experience versus a native app (other than the installation experience, and the lack of notifications on iOS)? We are really curious to see what people actually think after they try the app. If there really are issues that aren't fixable in a PWA, we'll do a native app too -- but so far we haven't really heard complaints that are limitations due to the fact that we implmented as a PWA. It is certainly true that the installation experience could be better, as having to separately install the telegram app is definitely not ideal.

Re: Suppressed alerts ...

Correct -- it is possible to determine whether not the notification was suppressed. That's what we've asked Rob about.

All motion is constantly analyzed and what was found is recorded. After that, there's a decision about whether the alert is "interesting" or not (based on feedback the user has provided) and another decision, based on frequency of occurrence, as to whether the alert should be automatically suppressed even if interesting.

When you review alerts in the UI, the defaults show you all "interesting" alerts. In the version Rob tested, we had just enabled the display of "interesting but suppressed" alerts by default as well, but there was a bug in it so I'm not sure whether or not the alert, if suppressed, would have shown up at the time Rob captured the experience.

We also have controls that allow you to see more, including "uninteresting" alerts, or even all detected motion w/o detected objects. They can also allow you to see less, by filtering the alerts by what objects were detected. The controls appear if you click the alert bell you see in the overlay.

Re: Xnor.ai

No, we are not using anything from them, and our knowledge about them is mostly limited to our experience with Wyze cam. They are impressive in the low amount of resources they require, especially given that they can do something useful on the Wyze cam hardware. On our Wyze cams, our own person detection detects people more reliably (on the same video stream), but theirs is pretty good too. We have yet to see ours miss a person that they detected, but we fairly regularly find examples we caught and they did not. I imagine they may be able to do a better job on more powerful hardware.

Sorry, we have never tried a thermal stream. I don't want to hazard any guesses about whether or not our model is appropriate for a thermal stream, but if you're aware of an accessible feed anywhere, we could hook it up and find out.

All analytics are running locally on the box. Your video never leaves the box unless you ask for it to, e.g. when remotely viewing, or because you ask for a copy of alerts to be backed up to storage like Dropbox or Google Drive.

We developed the analytics in-house, built on top of some widely available tools (e.g. Tensorflow, SSD, MobileNet) but with a lot of optimization to make them work well for home security usage on an affordable piece of hardware.

Yes. The device has a small internal SSD that holds the OS.

The disk compartment is also completely detachable (by moving two sliding clips) from the rest of the unit so that in the event of a failure a privacy-conscious customer can hold onto their drive compartment and send the rest back to us.