Different sensitivity for movement/AI at night vs day?
I have, for the most part, got daytime (well lit) movement sensitivity & human AI triggers working ok on 5 cameras (4 x cheap 1080p bullets, 1 x Dahua 5k dome).
At night the 4 cheap 1080p cameras with their ring of IR leds provide enough light for triggering.
But, the Dahua 5k dome is lit by street light a bit & doesn't go from colour to nighttime IR mode 'til quite late in the evening. I can set it on a timed schedule, but haven't explored that much yet.
Its grainy colour doesn't trigger on movement well at night in non-IR mode, & the AI doesn't detect humans well. Even when IR kicks in, its trigger sensitivity is way different to in daylight, possibly due to it having only a single IR led (pleasantly subtle, if less-effective, compared to the bullet cameras' "angry red ring").
While I try to sort its IR turn-on time it occurred to me that it may be nice to be able to set different trigger values for daytime vs night/IR, on a per-camera basis.
Before I actually request a s/w enhancement request, what do people think? I know I could add external IR lights off-camera, but just wonder if dual day/night sensitivity is a good idea.
TIA for your thoughts...
Comments
-
BTW, using 5.4.0b5 - daytime human AI has def gotten better with these beta releases. Now just the nighttime stuff.
Was detecting a lot of cats if movement sensitivity set too high. Raised human AI threshold & lowered movement sens 'til cats stop triggering. Not an easy balance to find...
-
From your description, it seems that the night image from this camera is very noisy when in day mode lit by street lighting - is this the case?
SecuritySpy's first-line (non-AI) motion detection works by comparing changes in pixels values between frames. In good lighting, the amount of pixel value changes caused by noise is tiny compared with the amount of pixel value changes caused by real motion, so it's possible to determine real motion with a high degree of accuracy. But, the more noisy the image gets, the more difficult it is to distinguish noise from real motion, until in extreme cases, the noise dominates and it's impossible to detect real motion. Adjustments to sensitivity may help a bit, but really the root problem is the noise, and fixing this is the best course of action.
The story is similar for the AI object detection - noisy images will reduce the accuracy of this significantly. Also, cameras tend to increase their exposure time in low light, causing objects to be blurred, which further decreases accuracy.
As you correctly state, improvements could be achieved by adjusting the camera's IR settings so that it switches to night mode earlier (most will have an adjustable threshold to set at which light level they will switch to night mode) and/or installing separate IR lighting that comes on at night.
If you want further input on this from us, please post an image taken from the camera at night so that we can see the kind of quality you are getting (to save an image, right-click on the camera's video feed in SecuritySpy and select the Save Image option).
-
Thanks Ben. Yes, image is grainy if IR hasn't switched in. I'm trying 2 things:
- Earlier cut-in time for IR (timed schedule vs Auto), &
- A Uniview camera, same 5Mpixel res, but apparently with better low-light performance than the Dahua (which I'll redeploy elsewhere). Note: Uniview may not support Bonjour, so getting it recognised by SS has me stumped for now - "Error 998 (see log)". Separate support email for that one if it defeats me!
-
Can't the day and night motion detection sensitivity be changed with an AppleScript running on a schedule in the App?
-
Yes, you could run an AppleScript regularly to set settings like this, but this is a little awkward to set up, and also it doesn't solve the fundamental problem that motion/object detection just can't work well with a bad image.
-
An update with major success replacing Dahua IPC-HDBW5541R-S (fixed 2.8mm lens 5Mp camera) with new Uniview IPC3235SB-ADZK-10 (varifocal 5Mp camera)!
Randomly altering camera configs in UniView web interface - by default the camera (as does Dahua) enables 3 streams - primary, sub (720p) & 3rd stream (D1) by default. With 3 streams, both cams work in their respective web softwares. Dahua works with all 3 streams in SS, UniView gave Error 998.
I disabled theUniView's 3rd stream & selected 2880x1620@25fps setting for primary, as I needed wide field of view & high resolution with good low-light sensitivity. All of a sudden SS connected & it works! Error 998 must be due to an incompatibility with some aspect of the 3rd (D1) stream.
Running H265 in camera's web config s/w, but SS in Stream Autodetect reports 'media_profile1 - H264 - 2880x1620' & 'media_profile2 - H264 - 1280x720'.
Dahua cam also showed as H264 in SS despite being set for H265 in its web interface. Don't understand why they both show H264 when set to H265, but don't care much. As long as file sizes are manageable.
Anyway, running 10frames/sec with high kb/s VBR with highest quality setting gives high detail & smooth movement without huge file sizes.
Uniview's clarity, detail, focus & low-light performance is much better than Dahua, despite being a cheaper product to buy (RRP of each is ~$260AUD, but I paid less).
I can see leaves waving around in the breeze, in focus, at 3m; blades of grass at 5m. Nightime movement, clarity & AI detection is still lower than daylight, but that was always going to be the case. Clarity under streetlights is great (compared to Dahua). Let's see how many cats I record tonight!
In a few days I'll have it fine tuned a bit better, & may be able to do a side-by-side comparison with the Dahua (time permitting).
I'll try to determine the exact cause of the 'Error 998' to help the general knowledge base, but not a high priority ATM.
Running scripts to change sensitivity not going to happen. I'd rather just run a light on a timer to give the camera more light to work with.
