New RCA HSDB2A 3MP Doorbell IP Camera
- Thread starter fleece
- Start date
flynreelow
Known around here
- Dec 12, 2016
- 1,349
- 1,235
Hello all. Anyone uses any kind of face recognition of this wide angle doorbell cameras?
What I wish is using Home Assistant to detect who is at front door. My doorbell is Ezviz DB1. Currently installed on my desk where I will first make a "lab" wiring with also smart RFID reader (ESP-RFID). Using MeanWell 12V DC power supply doorbell works fine now for days. I will use it to power my ESP-RFID board, doorbell and door lock. I tried many face recognition options for Home Assistant but one that stands out and seems to work (without cloud) is faceapi-js. Google found a nice partial implementation solution from this url:
www.wouterbulten.nl
What I find most problematic is the wide angle image this doorbell cameras provide. All faces are "round" and it is hard to detected if face is mine or wives. Needs to be really clear condition.
When front door motion is detected (Digoo RF PIR to Sonoff RF bridge using Tasmota and MQTT; I do not take motion event from doorbell as I do not know how to get it to HA) I call an url to the server running faceapi-js daemon script from URL above. It will take snapshot from camera (mjpeg feed from Zoneminder where I get reliable RTSP stream) and analyse it. All detected faces will be sent in json string to a Home Assistant Webhook automation. If unknow person is found it will do this if known person is found it would do that. All this is working find so far, just the face detection in wide angles is not so reliable.
Anyone using some other solutions that works fine also with wide lenses?
What I wish is using Home Assistant to detect who is at front door. My doorbell is Ezviz DB1. Currently installed on my desk where I will first make a "lab" wiring with also smart RFID reader (ESP-RFID). Using MeanWell 12V DC power supply doorbell works fine now for days. I will use it to power my ESP-RFID board, doorbell and door lock. I tried many face recognition options for Home Assistant but one that stands out and seems to work (without cloud) is faceapi-js. Google found a nice partial implementation solution from this url:
Local face recognition for Home Assistant using TensorFlow.js, Part 2: Learning to recognize faces | Wouter Bulten
Second post of my series on face recognition for presence detection in Home Assistant. In this post, I create a face recognition system.
What I find most problematic is the wide angle image this doorbell cameras provide. All faces are "round" and it is hard to detected if face is mine or wives. Needs to be really clear condition.
When front door motion is detected (Digoo RF PIR to Sonoff RF bridge using Tasmota and MQTT; I do not take motion event from doorbell as I do not know how to get it to HA) I call an url to the server running faceapi-js daemon script from URL above. It will take snapshot from camera (mjpeg feed from Zoneminder where I get reliable RTSP stream) and analyse it. All detected faces will be sent in json string to a Home Assistant Webhook automation. If unknow person is found it will do this if known person is found it would do that. All this is working find so far, just the face detection in wide angles is not so reliable.
Anyone using some other solutions that works fine also with wide lenses?
Last edited:
Hello all. Anyone uses any kind of face recognition of this wide angle doorbell cameras?
What I wish is using Home Assistant to detect who is at front door. My doorbell is Ezviz DB1. Currently installed on my desk where I will first make a "lab" wiring with also smart RFID reader (ESP-RFID). Using MeanWell 12V DC power supply doorbell works fine now for days. I will use it to power my ESP-RFID board, doorbell and door lock. I tried many face recognition options for Home Assistant but one that stands out and seems to work (without cloud) is faceapi-js. Google found a nice partial implementation solution from this url:
Local face recognition for Home Assistant using TensorFlow.js, Part 2: Learning to recognize faces | Wouter Bulten
Second post of my series on face recognition for presence detection in Home Assistant. In this post, I create a face recognition system.
What I find most problematic is the wide angle image this doorbell cameras provide. All faces are "round" and it is hard to detected if face is mine or wives. Needs to be really clear condition.
View attachment 76579
When front door motion is detected (Digoo RF PIR to Sonoff RF bridge using Tasmota and MQTT; I do not take motion event from doorbell as I do not know how to get it to HA) I call an url to the server running faceapi-js daemon script from URL above. It will take snapshot from camera (mjpeg feed from Zoneminder where I get reliable RTSP stream) and analyse it. All detected faces will be sent in json string to a Home Assistant Webhook automation. If unknow person is found it will do this if known person is found it would do that. All this is working find so far, just the face detection in wide angles is not so reliable.
Anyone using some other solutions that works fine also with wide lenses?
I don't use HA, but you can try ONVIF to get the motion event out. I think there is also an MQTT to ONVIF bridge somewhere.
You could try using dewarp filters in ffmpeg, but I suspect it would be too slow for facial recognition on a door lock. Otherwise you might want to mask out the capture to only the center of the frame where there's less distortion
EDIT: Just tried the dewarping option in FFMPG based on a stackexchange post I found. Here are a few parameters to try to get you started:
Pull single jpg and convert to dewarped jpg
Code:
ffmpeg.exe -i .\warped.jpg -vf "lenscorrection=cx=0.5:cy=0.5:k1=-0.227:k2=-0.022" dewarped.jpgPull RTSP stream and convert to GIF
Code:
ffmpeg.exe -i "rtsp:/rtspserverurl" -vf "lenscorrection=cx=0.5:cy=0.5:k1=-0.227:k2=-0.022" dewarped.gif
Last edited:
This actually works fast. Gonna play with this and see how much better will detection work. Thank you!
I don't remember which version allowed OutDoor Bell to be turned On/Off, but think that was the main change. If you don't need this I would stay with your current firmware. If you try the new firmware, you could always go back, I attached 191211 for you.
@silencery Here is a good reason to keep old firmwares.
Have you played with any settings in Batch Config. Tool? Go to CCD section, you will have fun there
Also play with WDR, looks like you have background washout...
We need to remember these Cams are to capture someone at your Door, not at the street.
HTH
I would say a Hard Reset, also kill power to it for more about an hour, when you see no more lights. But here is RCA firmware Build 190124 from Jan. 30, 2019, only one I have of RCA...Anyway, I would install this firmware First then do the Reset.
Good point, this is a Fish Eye Lens so Facial Recon may be harder...Hello all. Anyone uses any kind of face recognition of this wide angle doorbell cameras?
What I wish is using Home Assistant to detect who is at front door. My doorbell is Ezviz DB1. Currently installed on my desk where I will first make a "lab" wiring with also smart RFID reader (ESP-RFID). Using MeanWell 12V DC power supply doorbell works fine now for days. I will use it to power my ESP-RFID board, doorbell and door lock. I tried many face recognition options for Home Assistant but one that stands out and seems to work (without cloud) is faceapi-js. Google found a nice partial implementation solution from this url:
Local face recognition for Home Assistant using TensorFlow.js, Part 2: Learning to recognize faces | Wouter Bulten
Second post of my series on face recognition for presence detection in Home Assistant. In this post, I create a face recognition system.
What I find most problematic is the wide angle image this doorbell cameras provide. All faces are "round" and it is hard to detected if face is mine or wives. Needs to be really clear condition.
View attachment 76579
When front door motion is detected (Digoo RF PIR to Sonoff RF bridge using Tasmota and MQTT; I do not take motion event from doorbell as I do not know how to get it to HA) I call an url to the server running faceapi-js daemon script from URL above. It will take snapshot from camera (mjpeg feed from Zoneminder where I get reliable RTSP stream) and analyse it. All detected faces will be sent in json string to a Home Assistant Webhook automation. If unknow person is found it will do this if known person is found it would do that. All this is working find so far, just the face detection in wide angles is not so reliable.
Anyone using some other solutions that works fine also with wide lenses?
Do you have Blue Iris? Here is a very active Thread:
[tool] [tutorial] Free AI Person Detection for Blue Iris
This program analyzes motion in Blue Iris cameras in real-time using Artificial Intelligence. It notifies you if a human or an object of your choice is detected and can send alert images to Telegram. It is based on DeepQuestAI, a free multipurpose AI that you can install on your own server...
Sorry, only help I can give, I don't even have HA yet
flynreelow
Known around here
- Dec 12, 2016
- 1,349
- 1,235
Easiest way to install the firmware. I think I put it on th SD card?
I don’t know about fw 191211, but my EZVIZ-DB1 came with 190304. I upgraded to 200904 three days ago without any issues. I also upgraded my 16v-10w transformer to 16v-30w before my install. So far, so good.
Personally I would use Batch Config. Tool.
But only thing I am thinking is your Doorbell is registered on the Cloud. I remember Pete saying he bought a used DB and since it was already pre-registered he could not setup. Wonder what is the best way to do this? Worse case you give the new owners your Cloud account, you could give it a new password. Hopefully others will Chime in.
Good thing you upgraded, we were having issues with 190304 awhile back...It is a very old firmware...
Tried it today with light turned on in room. Sadly the detection points are worse. I will retrain it when light is better and see if it makes difference. But ffmpeg to "straighten" the image works great and fast. I am still using snapshot of the mpjeg stream that Zoneminder produces from RTSP. It is very fast. Faster then any other detection software.
ChappyEight
n3wb
I just picked up the Nelly's NSC-DB2 and installed it today. Got it connected to the network and activated/running in the Guarding Vision app without issue. I've assigned the doorbell a static IP through my Unifi router and the signal is strong. However, I cannot get the camera to be found in Blue Iris. When running the Batch tool, it verifies what I'm already seeing in my router menu which is that I do have the correct IP address. Also, I am able to see the camera at the hik-connect website. Finally, I am able to ping the camera from other devices on my network.
I'm hoping someone might be able to lend a hand to me because it seems most people have no issues connecting this camera to Blue Iris.
Thanks,
Chappy
I'm hoping someone might be able to lend a hand to me because it seems most people have no issues connecting this camera to Blue Iris.
Thanks,
Chappy
Which firmware version/build are you running? The latest Nelly's firmware supports ONVIF so all you should have to do is Find/Inspect
Here are my BI settings. Password is the Verification Code on the back of your Doorbell.
/Streaming/Channels/1?transportmode=unicast&profile=Profile_1
Check out our Doorbell 101 for tips...
ChappyEight
n3wb
So, embarrassingly, I was adding "softap_" at the beginning of the verification code because it looked like in the manual that's what I was supposed to do. I just did the verification code only and I was able to connect to the video. Sometimes it's the simplest things that spark it in my brain, like when you simply said that the password was the verification code (it clicked in my brain to try that alone).
Thanks for the help.
Thanks for the help.