Nice integration! When you click on the picture glance or picture entity, is it supposed to show nothing for default action?
I have the “https://github.com/fwestenberg/reolink_dev” integration installed, and my camera (410) is showing up on the main overview dashboard. Everything seems to be working fine. The only thing Im not sure about, is should clicking on the lovelace actually open a live stream up?
I don’t have the stream: component added to my config, should I? Thanks
1 Like
I have mine setup on my overview using a picture glance card. Right below where you pick the camera entity is the camera view. Change that from auto to live and now your card will show a live feed.
I also have it setup to open a large window when I tap on it.
On your overview dashboard, create a new view. Set it to panel mode.
On your new view add a picture glance card.
Select your camera and camera view should be live.
Set Tap action to navigation.
Navigation path should be your home view or the view that has the original picture glance card .
Now you need to go back to your home view where your original picture glance card is.
Change the camera picture glance card’s tap action to navigate.
The navigation path should be the new camera view you just created.
Now, when you tap the small card image, it will open up to a full screen live view. Tap the full screen image and be taken back to your dashboard.
2 Likes
So I configured an http to https proxy to expose the Hass over port 80 in order to get the motion sensing subscribed properly, but looking at the source code, it is using the URL info from the Hass configuration to set the subscription up. Is there anyway to override the port information for the motion subscription? I am able to login to Hass over both http on port 80 and https on port 8123.
While logged in with port 80, the integration fully works
From a security perspective my plan is to limit the http to https proxy to only be accessed from the nvr. I currently have it fully open, and Hass is functional on both ports.
Perhaps someone who is more python savvy than me have an idea of how I can ‘hack’ this in? Or could we modify this integration to include an option for subscription port override? Any help would be much appreciated.
Also, if anyone needs my nginx proxy config for their use, please let me know.
Thanks!
1 Like
Has anyone used this integration with the new people / vehicle detection cameras? If so, how good is the motion detection in Home Assistant? Do you still get false positives or does HA use the person / vehicle alerts to trigger the motion binary sensor?
I have a 410, decided to completely disable the internal motion detection due to all the false triggers.
I now have it integrated with Local realtime person detection for RTSP cameras
And now I only get a motion event in home assistant when it actually “sees” a person, not a moving tree, bugs flying around my lights, the community rabbits, or a bird flying by. I highly recommend this if you have the ability to run a docker container.
1 Like
Also with scan interwal 1-2 seconds it looses lot of motions…
3 Likes
Hi, with the Reolink RLC-810A with AI in the camera:
- Is it possible to use that in combination with HA to get a notification when people are detected without any processing (Deepstack etc.)?
- Can these video clips be uploaded automatically to Nextcloud with for instance FTP? (Initiated by HA?)
Thanks!
1 Like
FYI - if it seems like this isn’t working, you can’t have the Reolink software running and have this motion detection sensor work. Fought with that for about 30 minutes before I closed the Reolink app and then the sensor started reporting motion. Go figure.
What’s the URL rs=wuuPhkmUCeI9WG7C variable for?
Noy yet I’m afraid. Some users at my Github library did post some workarounds, please check it out at the issues.
1 Like
I am planning to buy a number of RLC-510A/RLC-810A (i.e. the models with person/car detection built-in) in order to use them in combination with my alarm system, which made itself with Home assistant (with “alarmo” custom component). Basically I’d like to have a binary sensor (mqtt or whatever) that will be triggered once a “person detection” is triggered by any camera. In this way I can receive notifications only when the alarm system is ARMED.
My question is: with the custom component “fwestenberg/reolink_dev” will I be able to do that?
My HA server is local, but I don’t have the NVR (or at least I am not planning to buy the reolink one but build my own with a rpi/rockpi).
Thanks, made a template sensor and works fine. Have you any idea it’s possible to use the recorded video playback and see the amount of space used/left from the sdcard. Thx
Everyone asking about person detection have a look at this open issue on the custom integration GitHub.
To summarize, the Reolink API doesn’t currently support exposing this, but, their engineers are working on it. So I guess y’all need to be patient 
Can anyone share any experiences with the built-in AI detection accuracy of these cameras vs something like Frigate? Personally, I think Frigate is awesome but it’s a resource hog for low power systems even with a Coral TPU. If the built-in AI of the Reolink cameras is as good then it’s a massive saving in terms of hardware cost.
1 Like
many thanks for the clarification! Actually in my case, this issue is blocking me from buying these cameras. At least as soon as I see the firmware update with the person detection implemented with ONVIF.
I don’t want to be too pessimistic, but here it might not be only a technical issue. Actually the firmware modification should be very easy for Reolink engineers, whereas there might be a strategy behind this lack (so you have to use their NVR). And honestly it is also fair from their point of you.
My vision about objects detection (person detection is my priority) is that there are many interesting projects (Frigate, Shinobi, DOODS) but none of them could be safely used for triggering an alarm with sirens (or in general to have reliable notifications), due to the presence of false positive. BlueIris is still one step above, but it not free and only for Windows. This is my perception in the first half of 2021, but I think that in the long run (some years) some of these project will outperform the current local detection of Reolink-AI cameras (which is good according to some youtube review I’ve seen).
So my idea was to use this local object detection for some time, and eventually switch to one of those project in the future.
Using the local detection would have been a saving in terms of : time (spent to configure these sw), hardware and power consuption…
Let’s keep the finger crossed! 
thanks
I run Frigate and I get the occasional false positive but its not very often. I wouldn’t have any sirens linked to any detection method unless it was 100% reliable. Contact sensors on doors seem to be the best bet for that and I have my security automation linked into those. I don’t think any AI is at that level of accuracy yet or will it be for some time but I might be wrong. Still, it’s a LOT better than any motion sensing I’ve tried. If you want to trigger alarms from object detection, maybe try number of detentions per x seconds. That’s what I’m going to try for mine. In my experience false positives aren’t repeated in low motion areas (indoors). I would always test with notifications first and then work your way up to a big alarm automation!
Yeah, youtube isn’t great when half the “reviews” are product endorsements so I was looking for some more honest opinions. 
yes, I agree , but I’ve seen a number of reviews and especially with daylight the person detection seems to good enought.
I do have the same bad experience with external PIR sensor: even with the ones that are supposed to be “PET-friendly”. PIR sensors for indoor and door sensors (magnetic reed) are for sure quite reliable… but obviously the burglar is already inside the house…
Many thanks for sharing your experience with Frigate: number of detections in a certain time is a great tip to prevent false positive. I’ve seen that with DOODS you have also the % of detection confidence, so you can sum those values in a certain period and set a threshold.
BTW: in how many cameras are enabling the person detect? which HW are you using?
I’m running 6 cameras:
1 x Reolink RLC-520:
Detect: 2048x1536, 30fps, Max bitrate 256kbps (5 fps in frigate)
clips: 2048x1536 30fps 6144kbps max bitrate.
Video Codec: H264 - MPEG-4 AVC (part 10) (h264)
Audio Codec: MPEG AAC Audio (mp4a) 32khz
3 x TPLINK Tapo C200
1 x Detect: 640x360 5fps
2 x Detect: 1920x1080 5fps
Clips: 1920x1080 15fps
Codec: H264 - MPEG-4 AVC (part 10) (h264)
Audio Codec: PCM ALAW (alaw) Mono 8khz
2 x Neos (wyze clone) running dafang firmware:
clips/detect: 1280x720 15fps 300kbps (5fps in frigate).
Video Codec: H264 - MPEG-4 AVC (part 10) (h264)
Audio: none
I was using a RPi4 but I wanted to use internal M.2 SSD and a Coral USB but power issues scuppered that. I’ve switched to a small inexpensive NUC with an Intel N4000 CPU now. Sits about 60-80%. Have a look at the frigate mega thread if you want to read more.
You can do the same in Frigate and also adjust the min size and max size of the objects which helps quite a bit. I find outside detection accuracy challenging, it’s a constantly changing environment. I have clouds reflected in my car back window detected as a person the other day 
True, but I guess it depends on each use case. I don’t think in my case the time between a camera detecting an intruder outside and them breaking a window and getting in, would be much different. From a security perspective cameras are more a deterrent than intrusion sensors too. If they are serious at breaking in though, then disabling cameras isn’t hard either.