Intrusion detection
I'm curious if Ben, or others, have views on the practicality of using network cameras and SecuritySpy as an alternative to traditional alarm systems for intrusion detection.
Even with a great alarm company I've found the latter to be problematic. In our case cohabiting with greyhounds and cats seems to defeat the older technology.
We long had motion detectors, which caused frequent false alarms from the dogs (despite supposedly having tech to prevent that). Now we have glass break detectors, which also cause frequent false alarms due to high pitched barking and rooing, no matter how they are configured and located. Short of alarming every window, screen, and door, which has its own problems, we have no good solution. Unless someone out there is interested in three big, adorable dogs 😎.
Now that AI person detection has improved so much would SecuritySpy potentially provide a better way? I imagine this would depend greatly on the number, location, and quality of the cameras. I was thinking of something like the 4 MP Dahua turrets with Night Color+ full color night vision, strategically located both indoors and outside.
Thanks, Bill
Comments
Hi Bill,
SecuritySpy does have very accurate human and animal detection. And, the better quality your cameras and lighting, the higher the detection accuracy that you will see.
4 MP is good, but 8 MP is preferred. And make sure to consider lighting as well. Even if the camera has its own infra-red LEDs for night vision, the quality of night images will be vastly improved with some supplemental lighting of the scene (either infra-red or visible). Also, consider the quality of the camera's optical system - auto-iris / auto-focus lenses provide better image quality than fixed lenses. More advice can be found at Achieving Effective Motion Detection.
No algorithm is perfect, and you will sometimes see humans classified as animals and vice versa (in perhaps 1 out of 100 events or so). This is because humans and animals have many of the same features (limbs, eyes, ears etc.) and with certain lighting/angles, they are difficult for an algorithm to distinguish. But, one advantage of SecuritySpy over a traditional alarm system that mitigates this issue is that you immediately get a notification on your phone with an image, and you can see at a glance what triggered the detection. And, as we expand our training dataset used to create our AI, the detection accuracy will only improve going forwards.
Thanks, Ben. I thought of the 4PM Dahua I mentioned because with Night Color+ it supposedly can record in full color, without supplemental lighting, in near total darkness. I assumed that full color would greatly aid the AI. But as you say higher resolution and more sophisticated optics are important, too.
For purposes of AI detection, can you say in general how much full color night vision helps? If not there would be many more camera options with other benefits.
Colour helps a bit with detection accuracy, but not very much. Sharpness and resolution are much more important than colour, so this should be your priority when choosing a camera and lighting. If the camera produces colour images, but they are grainy or they have a lot of motion blur (because long exposures need to be used), then this will give poor results. I would be surprised if any camera were able to produce good colour images at night at near total darkness - this sounds like an impossibility. Personally, I would go for an 8 MP camera that has good low-light performance but that is also sensitive to IR, and I would install some supplemental IR lighting to boost the quality of night images.