We are now seeing the beginning of what I think will be a very rapid shift to at least having reliable people detection analytics.
The hardware (whether it is Hisilicon here, Ambarella, Intel Movidius, etc.) is now getting here and the software too (e.g., XNOR on Wyze). It will not be rocket science to add such fundamental analytics to any camera.
The question is how fast will various camera manufacturers deliver. Dahua generally loses to Hikvision, in our testing but this is a big win for Dahua and an embarrassment to Hikvision's years of deep learning hype.
And Axis cannot be forgiven here. They sell a premium-priced product and while their VMD4 is strong for 'VMD', Dahua analytics+ clearly beats that at a lower price. I remember the whole Axis is late to megapixel in the late 2000s and then they swept in late and knocked out their competitors who got their early. Now, maybe history repeats itself for Axis here but maybe it does not and if Axis is late on analytics, it could be very significant pressure on them, as reliable people detection, once widely available, will become a must-have.
It's now almost 5 months since this review and I don't see any other viable low cost options for "reliable people detection analytics" from other brands. (Quotes do not indicate skepticism of the point in this instance. :-) ) I can vouch that the Hik Smart Series, even with extensive tweaking/trial and error of the settings, cannot come close to matching what is described in this report. Given the deal breaker limitations of Wyze AI, it looks like the Dahua is it on a limited budget.
Interestingly, it appears that if you don't want/need varifocal, that Dahua offers an even lower cost option with Analytics+. The N45DM62 is available for $199 at B&H, and it appears that unlike Hikvision that their fixed focal length option otherwise offers the same feature set. If anyone is familiar with this model, or can otherwise confirm it offers the same reliable people detection, I would appreciate hearing about it.
While I have a strong preference for 4K resolution at this point, the frequency of false alerts on my existing cams is such that I no longer have emails sent from them when a "smart event" is triggered. They simply record the video. It occurs to me that I could use a dual camera setup with a Performance series Hik with 4K (~$325) together with the 4MP Dahua at $200 and thus have reliable people detection and 4K (assuming the Hik doesn't miss much, which seems to be true most of the time) at about $500 total. Get a notice from the Dahua, then check both cameras recordings at that time. The extra camera could also be placed a small distance away such that it offers a slightly different perspective of the subject.
It would be very nice to see the image quality tests right here in this review along with AI results, even though they may be found in some of the shootouts.
One point on Hikvision, also: We have not yet tested AcuSense, but we tested their more expensive DeepinView camera, which did not perform as well as Dahua Analytics+, with more false alerts on animals, shadows, foliage, etc. Have you used both AcuSense and DeepinView? What differences did you see, if so?
I own one of the AcuSense cameras and it isn't even better than an older 2 series camera at false alerts. I own the DS-2CD2746G1-IZS and its analytics are being compared to an older DS-2CD2142FWD-I that I've owned for years. They both give false alerts at all times of the day or night. I'm talking about currently in clear weather and daylight.
On the other hand, my Samsung Smart Doorbell not only classifies humans, but also very reliably does facial recognition.
It would be useful for Dahua posts that the Dahua international PN is listed, while I understand you list the Dahua USA PN because that is the channel you are in, the rest of the world operates from the normal Dahua international part number system.
The best answer I can give to that is maybe. Dahua does not have a cross reference for US to international part numbers, at least publicly. It looks like it's the IPC-HDBW5442E-ZHE, but we cannot be sure that nothing has changed and that it doesn't have a slightly different processor, imager, or other component. I've reached out to Dahua to see if they can cross reference.
Dahua with its strong reinvestment on R&D would allow them to further their AI and implement them as standalone features in their devices.
As long as they focus on continuous improvement, ensuring the highest security for their end user, and offering value rather than low pricing, they should come stand and make a difference in the market.
I have recommended and sold Dahua and will continue as long as they are moving forward with products that will always impress our customers and make their investment on our recommendation worth it.
Agreed. Without reading this article, I probably would not have heard of their new Analytics+ cameras. I've been woefully unimpressed by anything Hikvision has released over the past couple of years and was excited to read how well the onboard analytics performed on these models.
I just ordered the N45 4MP dome & bullet and am anxious to do some in-house testing. If all goes well, I may just stop selling Hik altogether. I've been exploring the updated iDMSS app and am glad to see you can filter events by type, which is something you won't find in almost any other mobile app.
I did notice, however, that there doesn't seem to be any way to enable push notifications based on a schedule. Does anybody know if it is possible to receive notifications on a schedule in the iDMSS app? I know you can have a notification schedule on the camera itself, but I don't want to disable the actual analytics event.
I've been trialing an OEM version of this technology and am impressed overall. One weakness appears to be the nighttime performance when the IR is engaged. Rain and insect activity have triggered the intrusion alarm with no adjustment able to mitigate. With the IR off and external lighting utilized the performance was flawless.
Short of keeping a supply of pet moths to release during the IR testing portion I don't know how you would simulate an insect scenario but a shower hose attachment and a basin beneath would simulate the kind off tropical downpour we see around these parts.
Is the issue rain dripping down the dome or just random alerts from the IR reflecting on heavy downpour? We didn't have any issues with rain during testing but I can try and recreate the issue here. As far as the bugs, that is a hard one to replicate in the winter around here.
Can you answer a couple other questions? Is there any option to set a post event record time? Any way to send more than one snapshot via each email? Any way to eliminate the unnecessary event ended email that has no snapshot?
How about the quality of low light video? Do you also see the heavy noise/ghosting?
That appears to be the case Rob. It's still not 100% accurate but I'm impressed. For context this is a parking lot view without foliage or grass. The scene is partially lit with a new white LED street light and I'm using approx 50% camera IR output. On the primary Motion Detection setup page I only have 'enable' ticked and the default area setup grid enabled. Smart Motion Detection is enabled and both options ticked with the middle sensitivity selected. The IVS rule is setup as intrusion with record and snapshot selected.
When I view the previous days results the only activity is from people and vehicles so far.
I am testing the Dahua Analytics+ NVR: DHI-NVR4216-16P-I.
It works quite well in terms of pushing the human/vehicle detection and perimeter protection to non-AI cameras. However, it sometimes misses humans, namely when someone is riding a skateboard or a bicycle on the sidewalk and is not in view for at least 2 seconds. Or sometimes, when more than one person is walking near each other.
According to the user manual for this Analytics+ NVR, here are key limitations:
220.127.116.11 IVS (General Behavior Analytics)
The IVS function environment shall meet the following requirements: The object total size shall not be more than 10% of the whole video. The object size on the video shall not be more than 10pixels*10 pixels. The abandoned object size shall be more than 15pixels*15 pixels (CIF resolution). The object width shall not be more than 1/3 of the video height and width. The recommended height is 10% of the video. The object and the background brightness different shall be more than 10 grey levels. The object shall remain on the video for more than 2 seconds. The moving distance is larger than its own width and shall not be smaller than 15pixels (CIF resolution). The surveillance environment shall not be too complicated. The IVS function is not suitable for the environment of too many objects or the changing light. The surveillance environment shall not contain glasses, reflection light from the ground, and water. Free of tree branches, shadow, mosquito and bugs. Do not use the IVS function in the backlight environment, avoid direct sunlight.
I did post a YouTube video in an earlier reply that shows how I subscribed to receiving these AI/IVS alerts on the mobile app. But in doing so, I noticed two issues since yesterday:
1) Why was human motion missed when I left the office around 3:30 p.m. yesterday with supplies in my hand? 04.01.2020-11.48.25
2) And numerous false motion around 6am today? It can’t be due to the change in light from black/white darkness to sunlight because the camera didn’t go into color until 6.36am. 04.01.2020-11.51.51
In looking at the NVR’s settings again, I am seeing that the line crossing was changed from “both ways” to the unsecured side to the secured side. I forgot that I made this change yesterday, so the tripwire did not activate because I was leaving and not entering. This now makes sense.
Working on the rest now. Tech support is recommending drawing bounding boxes for minimum and maximum size. So far, no luck. I'll update this post once we find a resolution.
Here is a new video which includes drawing the minimum and maximum sized bounding boxes for objects.
Additionally, the Dahua product manual for the 16ch AI NVR specifically states the following requirements in section 18.104.22.168 under General Behavior Analytics:
The IVS function environment shall meet the following requirements:
* The object total size shall not be more than 10% of the whole video.
* The object size on the video shall not be more than 10pixels*10 pixels. The abandoned
object size shall be more than 15pixels*15 pixels (CIF resolution). The object width shall not be more than 1/3 of the video height and width. The recommended height is 10% of the video.
* The object and the background brightness different shall be more than 10 grey levels.
* The object shall remain on the video for more than 2 seconds. The moving distance is larger than its own width and shall not be smaller than 15pixels (CIF resolution).
* The surveillance environment shall not be too complicated. The IVS function is not suitable for the environment of too many objects or the changing light.
* The surveillance environment shall not contain glasses, reflection light from the ground, and water. Free of tree branches, shadow, mosquito and bugs. Do not use the IVS function in the backlight environment, avoid direct sunlight.
And Dahua Tech Support adds, "Each rule can set 2 target filters (maximum size and minimum size), that is, when the passing target is smaller than the minimum target or larger than the maximum target, no alarm will be generated. The maximum size should not be smaller than the minimum size."