I’ve freed up an old NAS that has about 5.6 TB of storage. As a result of this I have been playing with my security cameras in the past week.
I have five Tapos and one Ring. I have motioneye running in docker and have been experimenting with masks and motion detection. The tapos work well will motioneye but the Ring is too locked down.
I live on a corner and in the last 18 months I have been asked twice by police if I had camera footage. At the time I was not recording any camera data.
So given my setup does HLB have any recommendations for recording settings?
I have the option for recording video and photographs. These can be continuous or motion detected or at a specific time interval.
Obviously the correct setting depends on what is in the field of view but I am thinking of going with motion detected (in the unmasked areas of the shot) and perhaps a still photograph taken at one minute intervals.
I also wondered if anyone had had any experience with the Community Camera Alliance? I received a flyer about this a few months ago but did not pay much attention at the time. I suppose it is a good thing but would be interested in HLB’s views.
In my area we look like we are tracking along much the same as previous years but it looks pretty “hot” just around the corner.
Added image ‘/home/tp/Downloads/1.jpg’
The image shows a person wearing dark clothing and gloves, appearing to be attempting to open a door on a residential building. Here are the key details:
Person:
The individual is dressed in dark clothing, including a hooded jacket and gloves, which could suggest they are trying to remain inconspicuous.
They are reaching up to the door handle or lock, possibly trying to gain entry.
Door:
The door is a white, multi-paneled glass door with a frosted glass design, allowing some visibility but obscuring details inside.
The door is part of a residential building, as indicated by the surrounding architecture.
Surroundings:
The setting appears to be a backyard or patio area with a wooden deck.
There are potted plants and a small garden visible, along with a white wall and a brick building in the background.
A metal chair is placed on the deck near the door.
Lighting and Atmosphere:
The lighting suggests it is daytime, with natural light illuminating the scene.
The overall atmosphere appears calm, but the person’s actions suggest an attempt at unauthorized entry.
The image could depict a scenario of someone trying to break into a house, though the exact intent is not clear from the image alone.
A very timely thread for me - I picked up a TP-Link Tapo C200 on sale last week. I’ve seen a few of the proprietary setups of friends and family and wanted something a little more open; Ideally completely open and not dependent on cloud services. Despite the requirement for app setup, the Tapo ecosystem seemed like a reasonable place to start a proof of concept and deciding which direction I want to go as the C200 specifically supports RSTP and ONVIF (I assume some or all other Tapo devices do as well). Intent was to start exploring a few concepts around home security cameras, NVRs, and maybe some computer vision (person/object identification) things with the raw video streams, as these are all totally new areas to me. A bit of a confession - it was an impulse buy and a good excuse for a new gadget as well…
Therefore, I can’t make any specific suggestions for settings or comments about the Community Camera Alliance, but a few of the “ingredients” below that I’ve stumbled across may help somewhere. Interested in how you proceed with your setup, or whether you’ve had any experience with the below too. Thanks for linking motioneye - I hadn’t seen that yet.
The Tapo to Home Assistant integration worked flawlessly for camera control, and I can also view the raw ONVIF streams via HA. I can’t see any obvious way to record or manage surveillance footage in HA. I confess I had done zero research (beyond: ONVIF support? Impulse buy!) and had hoped that HA would have something built in that I could throw hard drives at and press Go and it would all magically work. I’ve not done much looking into it since having the camera, but have stumbled across the below:
Blue Iris seems to be very well regarded for self hosted NVR, but as far as I can tell isn’t open source and is a stand alone commercial product.
I know it doesn’t directly answer your question, but perhaps one of the above (e.g., Frigate NVR) might help in the sense that it is actively able to detect objects and start/stop recording or take still images based on a person being detected?
a motion detector has detected motion outside the front door, a picture has been taken automatically, should the police be called ? this is the picture /home/tp/Downloads/3.jpg
The decision to call the police depends on the context and the specific situation. Here are some factors to consider when evaluating whether to call the police based on the image you provided:
Key Observations from the Image:
The Cat:
The image shows a cat sitting calmly in front of a door. Cats are often curious and may sit near doors for various reasons, such as waiting for someone to open the door or simply observing their surroundings.
The presence of a cat alone does not indicate any suspicious activity.
The Door:
The door is a standard residential door with a small pet door or window. There is no visible damage or forced entry.
The door appears to be securely closed, and there are no signs of tampering or unauthorized access.
Motion Detector Trigger:
The motion detector has detected motion, and a picture was taken automatically. However, the image shows no human activity or suspicious behavior.
Considerations:
Nature of Motion: The motion detected could have been caused by the cat, a bird, or other non-threatening activity. Motion detectors can sometimes be triggered by small movements or shadows.
Context of the Alert: If there are no other signs of suspicious activity (e.g., forced entry, noise, or other unusual behavior), the motion might not warrant immediate concern.
Local Policies: Some areas have specific guidelines for when to involve law enforcement. If in doubt, it’s always best to err on the side of caution.
Recommendations:
Assess the Situation:
Check the surrounding area for any signs of unauthorized entry or suspicious activity.
Look for any damage to the door, windows, or other parts of the property.
Call the Police If:
You see or suspect someone attempting to break in.
There is visible damage to the property.
You hear unusual noises or see someone acting suspiciously.
You are concerned for your safety.
If No Immediate Threat:
If the situation appears calm and there are no signs of a threat, you may choose not to call the police. However, you can still monitor the area or check the property more thoroughly.
Conclusion:
Based on the image alone, the presence of a cat and a closed, undamaged door does not indicate a clear threat. However, if you are uncertain or concerned, it is always best to contact the police or local authorities for further guidance. Safety is the top priority, so trust your instincts and take appropriate action.
Authors note:
Computer: AMD Ryzen 3600 with 64GB ram, single 1GB NVMe card.
GPU: Single Nvidia RTX3060 PCIe video card.
AI: qwen2.5vl:32b free by Alibaba, running locally under Ollama 7.0 on the above machine.
And finally : qwen2.5vl:72b
Note: 1.jpg above is “rear.jpg” (man in black) and 3.jpg is “front.jpg” (the cat)
you are a dedicated and alert security officer who does not guess if not totally positive.
Your job is to monitor pictures from two cameras on the building. ‘front’ camera faces a grey door with 3 small glass panes in its center. ‘Rear’ camera faces a glass enclosed area beyond a wooden deck with some pot-plants and steel chairs.
You will receive pictures from these cameras from time to time and you will analyse them for threats to the building, i.e. fire or intruders and animals that are harmless but will have tripped the motion sensors in the cameras and created the pictures you have received.
You will respond to all pictures with only three responses.
1: “ALERT INTRUDER - CALL POLICE”,
2: “ALARM MOTION DETECTED, CHECK AREA”,
3: “False alarm, no emergency” when a animal or bird was the cause. No other responses are required.
Understood. I will monitor the pictures from the ‘front’ and ‘rear’ cameras and respond with one of the three specified responses based on the analysis ofthe images.
Please provide the pictures for analysis.
Added image ‘/home/tp/front.jpg’ False alarm, no emergency
Notes: this time the AI was the same but in a larger image of 72b and 48GB in size. This nearly overwhelmed my poor PC as I only had 15GB ram spare when it was running.
qwen2.5vl:72b 05ea68274581 48 GB
Thanks @Belfry / @techman, adding AI to a home security setup sounds interesting and I might have a go at Frigate at some stage. However, I am not sure AI would help much for my current problem.
I am interested in capturing most of the footage that crosses into my unmasked zones or those events that generate a loud noise, such as a motor vehicle accident. Most of the triggered events will be benign and both humans and AI would have difficulty in sifting out the more interesting events.
For example this is the image and video from yesterday of the Amazon delivery guy throwing my new Tapo camera over my locked gate.
72b is enormous. Not tried any of the vision models myself yet, but it’s quite possible that 64GB isn’t enough by the time it loads the model (50GB+) and does actual processing too. What sort of graphics card or accelerator are you using in conjunction with the model? @techman’s extra 2x RTX3060s and the added VRAM are undoubtedly helping here. @techman - curious to see some figures from nvtop and top when you’re running that model!
Yeah it runs fine, single RTX3060, Ryzen 3600 cpu ($130 with fan) 64GB DDR4, that’s all.
tp@ubuntu:~$ ollama run qwen2.5vl:72b
what can you see here ? /home/tp/front.jpg
Added image ‘/home/tp/front.jpg’
The image shows a black cat sitting on a doorstep, facing a closed gray door. The door has a modern design with three rectangular glass panels arranged vertically. The cat appears to be looking up at the door, possibly waiting for someone to open it. The doorstep is made of a light-colored material, and the overall scene suggests a quiet, domestic setting.
PID USER DEV TYPE GPU GPU MEM CPU HOST MEM Command
105209 ollama 0 Compute 0% 9152MiB 74% 587% 40740MiB /usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-ae3823d63bf19077ec4c484843defaf89
