Something kinda cool I did with outdoor lights

[brad5]

I thought this was kind of cool so I thought I would share and see if anyone had ideas for improvement.

I recently installed Hue outdoor flood lights around the perimeter of the house - 9 in all. Then I installed outdoor motion sensors in each lighting "zone" - one zone for each side of the house. I used Hue motion sensors just because I have had success with them elsewhere. I have each zone configured so I can put them in "auto" mode - simply a switch that enables or disables activation in the "room lighting" app. I also have a restriction so the lights will come on only from sunset to sunrise-30. Works fine, but despite the motion sensors set on low sensitivity, they trigger fairly often. (If anyone has a better outdoor motion sensor, let me know!)

I also have Arlo cameras in all the same locations, and Arlo has a person recognition feature. I turned it on and set up IFTTT to activate a virtual HE switch whenever a camera detects a person. I then modified my room lighting app so instead of activating on a hue motion sensor it activates on a unique virtual switch called "activate lights" I created for each zone.

To summarize, each of four zones (north, south, east, west) has a number of virtual switches and a Hue motion sensor:

north Arlo detects person
north motion
north Auto activate on motion
north Auto activate on person
north Activate lights

I built a set of rules that turns on the "activate lights" if the Hue motion sensor activates AND the "auto activate on motion" is on, and a set of rules that turns on "activate lights" if "arlo detects motion" turns on AND "auto activate on person" is on. So for each zone I can activate on motion, on arlo detecting a person, on both, or none. In addition to the lights coming on, any person detection triggers an Echo Speaks alert, letting me know there's someone out there.

I would prefer a person detection system other than Arlo - too many points of failure. However, as Donald Rumsfeld famously said, "you go to war with the army you have, not the army you might want."

The system actually works remarkably well. There is a delay for person detection but it's not too long as to be unusable.

Here's the SharpTools dashboard I built to configure and monitor it all:

[moh]

I have used a very similar setup for the last two years except utilizing Camect for person recognition in conjunction with my outdoor PIR sensors. In addition to having it respond to Mode (sunset to sunrise), since I also have outdoor speakers installed (speaker output from a dedicated Echo Flex is connected small automotive amps that power the speakers), it also uses Echo Speaks to announce verbal warnings depending upon the arm state of my security system to potential intruders (have separate messages for Away Armed and Stay Armed warnings). This is in addition to my internal Echo Flexes and Dots and Echo Shows (and my phone through Pushover notifications) announcing whether anyone is detected outside and in which location. All of these features can be individually enabled or disabled (for trouble shooting or just WAF) with virtual switches. System has been very reliable (have not really had to turn anything off for trouble shooting in the last two years).

Also, since Camect does other types of object detection quite well (various types of animals etc), it can also announce not just if it is a person (in which a verbal warning can be enunciated via the outdoor speakers) but also let us know the type of animal that has come into the yard (sometimes we will get coyotes (dogs/foxes via Camect), deer or rarely mountain lions (cats) or more typically raccoons and squirrels in our yard (no bears in our neighborhood). The mountain lion one saved us once.

[brad5]

I like the speaker idea. I do have an Echo inside the front door that barks like a dog if motion is detected and we are not home. But I had not thought about verbal warnings. Some clips from the original robocop movie might be helpful.

Camect looks very cool. I am not a huge fan of Arlo and would rather go with something like Unifi cameras but just haven't made the move yet. Camect would be a very nice edition. I'll do some research.

[erktrek]

I feel the same way you do about Arlo - cloud based, api is problematic especially for non-subscribers. Unifi looks cool but is expensive. I've posted this before but here is a decent comparison of security cameras:

2022 Update BEST 4K PoE Security Camera: $1100 vs $110 Motorized Zoom Cameras.

Also battery operated ones (I have not watched this one yet):

Top 4 Wireless Battery Powered Security Cameras 2022.

[brad5]

Interesting results! I thought the empire tech cameras looked pretty good for the price. But their website is the most difficult garbage to navigate - I can't deal.

@moh what are you using for cameras?

[moh]

@brad5

what are you using for cameras?

I am ashamed to say (at least some people have issues with this brand) that I use 4K Amcrest turret cameras. They are relatively cheap (or at least reasonably priced depending upon your point of reference) and have excellent day time imaging. However, other more expensive brands (especially those that use the Sony Starlight sensors) have much better night/low light imaging. In general, 5MP cameras (and lower resolution) will actually have better night vision compared to 4K cameras due to pixel size. At any rate, since I have my outdoor lights either go from dimmed to 100% bright (during the evening hours), or go from OFF to 100% (in the dead of night), the mediocre IR vision is really not much of an issue in my use case. If anyone is interested, I will try to post some images (IR vs day) although this information is available in much more detail in comparison review articles on the web.

I like the turrets because they are much more resistant to vandalism, less prone to becoming misaligned than bullet style cams, and do not have the IR and condensation image reflection problem of dome cameras. But everyone’s use case and demands and therefore preferences will of course vary. The 4K and 5MP turrets are sturdily built, with all aluminum/metal alloy frames and camera housings (no plastics) at least on the models I purchased over the last two years.

Amcrest is also local (SoCal) to me so even when ordering direct or through Amazon, I usually get stuff from them in one day (I once got a camera delivered on the same day). Amcrest is not the highest end brand (not by any means) but they can also be flashed with the Dahua firmware for additional features if you are a fan of Dahua, which is made in the same factory. I have mine on their own local network with Camect and Hubitat so they cannot phone home to China. Overall, perhaps not the best of cameras (particularly for the resolution under night vision) but meets my needs. Hope this helps.

[moh]

Here’s a recent daylight capture (screen shot so not full resolution):

And a still capture of a mountain lion under night vision:

Image of the mountain lion is a little blurred as it was moving pretty rapidly and I attempted a screen capture from the video as I could not figure out how to post the actual video.

Edit: I should also note that since these are from my Camect downloads, they are at 1080P only in order to decrease the load on my Camect (a single Camect hub can only handle 24MP total active camera resolution currently although they are allegedly working on a higher capacity hub for larger applications). Full resolution is simultaneously captured on my concurrently running Amcrest NVR. I use the Camect for superior AI object recognition and use the Coral USB board plugged into the Camect to supposed enhance the AI processing speed of the Camect Hub. I use the recording from the Amcrest NVR in the event that LE needs a higher, full resolution recording.

[moh]

@brad5

I thought I would share and see if anyone had ideas for improvement.

In addition to the above enhancements, I forgot to add that since Camect also identifies various delivery services (USPS, Amazon, UPS, FedEx and DHL) via their logos on their vehicles, I also have it set up so that it announces which delivery trucks are making a delivery and which entrance (front porch or back door or mailbox) the delivery is being made to. There are so many things you can go down the rabbit hole with with all the object recognition in products such as Camect and Blue Iris when coupled with motion sensors and contact sensors through Hubitat. I am relatively new at all this as I have only been a user of Hubitat for just under two years now so still consider myself a noob, but am constantly finding new uses for all this stuff. Have fun with it!

[marktheknife]

@moh

And a still capture of a mountain lion under night vision:

The most dangerous animal I’ve managed to spot since I setup Camect is a raccoon!

I’ve been very impressed with Camect’s local AI for object recognition.

I was using Blue Iris with deepstack AI for object recognition but I find Camect easier to work with even if it’s less customizable/flexible than Blue Iris. @brianwilson’s Hubitat integration allows for enough automation for my needs.

[moh]

I heartily agree, thank goodness for @brianwilson!

[brad5]

Footage looks fine! I have about 10 cameras... at 5MP that's at least 2, if not 3, camects. I didnt realize you could split the feed and still have an NVR in the mix. Good to know.

I was headed down the Unifi route for cameras but the video review that @erktrek posted kind of convinced me it wasn't worth the extra money.

[moh]

@brad5

I didnt realize you could split the feed and still have an NVR in the mix.

Yes, but it is of course camera dependent (cam must be capable of multiple feeds) so you have to delve into the specs carefully. You probably know this already, but for those who are less experienced shopping for a new ethernet camera (like me not too long ago), before buying, it’s best to download a copy of the camera’s full manual if available online to get a more detailed idea of what the camera is capable of rather than just the manufacturer’s marketing spec sheet. The spec sheets will only tell you the various resolutions the cam is capable of, but often times not which resolutions are available for each stream. You may have to contact support directly for this info.

You can usually go into the settings of each camera by accessing through its LAN ip, then set the parameters (resolution, frame rates, bitrates) for each stream. If you are not utilizing a particular stream, you can disable it at the cam level to (I suppose) decrease the network load and the load on the camera itself. For those running multiple streams, you then just select (on your NVRs) the stream you would like to use for each cam on each NVR. As long as your network can handle the traffic, everythings good to go (at least in my experience, YMMV).

Most ethernet cams (I don’t use any WiFi cams so I can’t comment on those) I have looked at can do two, some even three simultaneous streams. Some of my older Amcrests actually can do three streams, while their “upgraded” newer models can only do two (cost cutting?). You have to check if they can do the secondary feed at a high enough desired resolution for your AI hub (in my case Camect) while still handling the primary feed (in my case to the NVR) at full resolution. In some cams, the secondary feed is limited to very low res, or the camera has trouble streaming at a full 4K simultaneously with a 1080P (or a 5MP secondary) stream. Luckily, all my cams are capable of at least a 1080P secondary stream.

Of course your network needs to be able to handle the load from all these simultaneous streams as well (you can adjust the bitrates and the frame rates of each stream to hit the “sweet spot” for your use case). I don’t recall the link but there are a number of camera sites that have published charts of minimal bitrates needed for any given resolution/frame rate combination. I’ve set my bitrates accordingly (just slightly higher) in order to minimize the load on my LAN.

[moh]

@brad5

I have about 10 cameras... at 5MP that's at least 2, if not 3, camects.

If your cameras can do a simultaneous stream (easy to check, just access the camera settings through the camera’s ip), then see what resolution is available on the secondary stream (which I send to Camect). I have found that on my Camect, that 1080p is more than adequate for the object recognition and for alerts, etc. With 10 cameras, that would be about 20MB total so you could accommodate all this on one Camect.

Another thing to remember is I don’t use Camect as the primary NVR, only for object recognition (for alerts and other automation through HE). The reason for this is that it only comes with a 1TB internal drive (I have added a 4TB USB drive to it) so it is also limited in how many days of recordings can be retained (dependent upon image resolution, frame rate, image complexity and movement etc). That’s the reason I also use a separate “conventional” NVR with a larger (12TB) hard drive that records at full cam res. Just another consideration.

[brad5]

@moh

Of course your network needs to be able to handle the load from all these simultaneous streams as well

My home network is rather robust. When I go the IP camera route I will add them to a separate camera VLAN, install a dedicated 1gb PoE switch, and connect it to my core switch via a 10gb link. If I keep the NVR/Camect on the same switch the traffic will stay local to the switch without ever traversing the rest of the network.

I was going to off until next year but you make it sound so cool I may have to accelerate the plan. The biggest expense is actually the labor to run all the cat6. I generally don't do it myself. An old house takes creativity!

[moh]

@brad5

@moh what are you using for cameras?

Had previously answered but just saw that the Amcrests were on “sale” for $100 (with applied coupon): https://www.amazon.com/Amcrest-3840x2160-Weatherproof-Microphone-IP8M-T2669EW-AI/dp/B08NXX27FR?tag=digiblurd-20

I think I paid $139 or $149 a year ago. I run main stream at 4k (8MP) to my main NVR, and simultaneously use Stream #3 at 1080P (2MP) to my Camect for object recognition. It also does Stream #2 at lower resolutions but I do not use this (I only use 2 out of the 3 available simultaneous streams).

Just wanted to address your simultaneous stream question if you were still interested. Hope this helps.

[brad5]

Yeah. I think it will be next year's project.

[rlithgow1]

I was looking at camect. I really liked the idea until it was pointed out to me that due to it's memory limitations, after 4 cameras, video resolution degrades considerably with each camera added. This was very disappointing. In the end I'm going to switch to Blue Iris (I currently use Xeoma)

[brad5]

Yeah that is a limitation for sure. But Blue Iris is a software-only solution, right? I'd really prefer a dedicated appliance. Besides, we're primarily a Mac household. I maintain a Windows VM to talk to my software-defined radio, but that's about it. I'm hoping by the time I'm ready to purchase they have a beefier version.

[rlithgow1]

I have several (just made 7) blade servers with several VM's each on the rack in my basement so I can throw it on one of those. I do like the appliance aspect but if I can't make out someone's face, there isn't a point to having a cam..

[brad5]

I'm gonna need a bigger rack...