camera detection / recognition research (what to buy and what is achievable)

Kabraxis

n3wb
Sep 3, 2024
1
0
Poland
Hi,

I am sharing what I have found so far and will continue my research with you. I do setup my first time. Please confirm / decline my conclusions so far and share your experience.

I am not interested in cloud solutions. I want to have server locally, fully independent. My own VPN on WireGuard. Home Assistant.

1. There is no solution on the market which let detect / recognize custom objects.
For example trash bin if system doesn't support trash bin already. It is possible to code in python with AI my own model (pure software developing AI work), but there is no solution where I can upload images of object and potentially click true / false to improve algorithm.

To summary this up the only solution is to do AI software developer work. No solution which let you achieve good result in simple way.

2. No advantage to have extra features in camera like motion detection, if system on server has motion detection. Let's say the only pros is moving CPU load from server to camera. Is it worth it? More overhead with settings for all cameras. Do I miss something? Maybe if buying Synology NAS which have strict limited in hardware, then it make sense.

3. System choice:

Do you recommend something else?

a) blue iris - install on Windows. This mean not negotiable NO for me.

b) Synology NAS (not DVA) + Frigate - maybe? I see downside here, but I don't know how important it is. I need place to store video anyway, so solution like Synology NAS is needed. Synology NAS has limited CPU and memory, so while choosing Frigate it seems to be reasonable to use separated server.

So I will have situation: cameras -> NAS -> Frigate, which mean all videos have to be stored in NAS and then transferred to Frigate sever which mean extra delay, network overload and usage of hard drives. How much doest it matter? What delay I can expect? What not obvious issues with that?

c) Synology NAS DVA - as I understand only Deep Learning NVR Series | Synology Inc. (DVA3221 and DVA1622) support recognition. I assume other versions support only detection (do they?).

It is not clear for me what this mean How many Face Recognition and Deep Video Analytics tasks can I use? - Synology Knowledge Center .

Can someone explain this?

Does it mean with 12 tasks I can have only 6 faces in database to recognize?

Does it mean I can recognize 6 faces at the same time for all camera, but can have 100 faces in database? What if there will be 10 people on cameras? 4 of them will be ignored or delayed?

I have no idea what this tasks mean, but it seems to be important factor to understand.

4. UniFi protect - I don't know too much about this one, but it looks like I kind of close into one brand. Especially for servers. It looks like a solution where buy hardware and easy to got, but with the downside of buying brand dedicated devices which limit choice in the feature. I can be wrong, but they have "cloud gateways" servers and everything seems to be around cloud which I am not interested in.



My setup in progress:

Everything on ethernet cables, category 6. Cameras on PoE.

systems:
- Frigate ? / Synology NAS ? / Synology NAS DVA ?

hardware:
  • Synology NAS for storage ?
  • extra server for detection / recognition ? recognition only? no extra server?

5x cameras:
  • 1x front camera to watch main entrance: ?
  • 4x camera around the house: ?
 
Last edited: