Dahua IPC-HDW5442T-ZE 4MP Varifocal Turret - Night Perfomance testing

Just wanted to note here as well that I've tested the SMD on these last 2 test releases and they do still have the issue of incorrect detection as @looney2ns also saw. In looking at the code it would appear that the last 2 test versions (7/31 & 8/14) are built off the 6/20 code which also had this issue hence it being inherited in these builds
it's not as if the bullets are doing great though the settings does not fit the eyeball cams.
I'm using those cams together with B5442E-ZE (the bullet) and I can tell you that the bullets are triggering cats and dogs (especially on IR light) as SMD events.
 
Have you tested MD with snapshots? I’ve found that the camera will not always place the date/time stamps/description overlays on the snapshots. Also, Anti-Dither isn’t working as intended. If you set the anti-dither period to 5s, the camera will continue taking snapshots through the 5s if motion is still being detecting instead of waiting 5s after the first motion detection which is what it’s supposed to do.
 
Last edited:
  • Like
Reactions: djernie
Have you tested MD with snapshots?
No, I didn't (yet) test it with snapshots.
As for the anti-dither, as I expected it to continue X seconds after detecting a motion. Only now I see that it's supposed to continue recording for a certain amount of time after motion has ended. I didn't even notice that with fast reading.

I'll need to check it (I believe that you're using the newest FW - 20200825 and I'm using the 20200814 FW) but I believe mine is also counting since the beginning of the motion. If that's the case, there's no reason for me not to upgrade as well.
 
  • Like
Reactions: biggen
According to the Dahua Wiki, the anti-dither setting "memorizes one motion event during the anti-dither". That in my mind means that if you have it set to 5s, then it will sleep for 5s after the first motion. What mine does is continue to fire snapshots during the entire 5s period.

I've set the Anti-dither 0s just to test to see if it was backwards from what the Wiki states and I honestly didn't see a difference. It continues to capture loads of snapshots even when set to 0s.

I'm actually running 2.800.0000000.19.R, Build Date: 2020-07-09 so I think I may be a couple FW behind you. I'll wait till the next baseline is released since these firmware you are testing are really dealing more with SmartIR than anything else.
 
  • Like
Reactions: bigredfish
All, I went out in the rain last night to test the latest firmware and have included a 14.5 min video below. Timestamps for ease of viewing are below.

Long story short:

Findings
  • SMD in my testing both day & night has been ‘improved’ since 8/14 BUTas @looney2ns and I agree, still has issues in identifying incorrect targets such as animals as humans (noted this to @EMPIRETECANDY last night). Therefore SMD still needs work so have sent feedback and video to Dahua in hopes to get this included in the September security baseline they are working on.
    • With regards to SMD sensitivity, there are some important takeaways here to dialing in your SMD targets for performance even while awaiting the next changes. I'll try not to lose people here as I explain what I've seen with SMD across these cameras. Specifically, sensitivity is usually understood as the higher the sensitivity, the more sensitive a given area is to an input or in this case a target BUT in the SMD implementation it appears that Dahua has taken Sensitivity as you would expect it to be implemented and combined this with 'threshold'. Take a look at regular Motion Detection settings and you will see it clearly has both Sensitivity & Threshold which are critical to any form of motion based capture. However, In SMD (although this is an effective add-on to regular Motion Detection) it is the threshold that it would appear the algorithm is using and biasing to spot targets rather than true sensitivity (as described). This means to enable more accurate captures or response from SMD we have to adjust our understanding and expectations of it a little. Specifically this means flipping the understanding so that setting to low 'sensitivity' (actually threshold) means it will trigger / find what the cam see's as targets more frequently than high sensitivity (again actually threshold) which would result in less target acquisition or a more conservative approach to identifying then tracking a target through FOV. I saw this in my testing and as I felt this was important, wanted to mention it here as it does change the way you approach SMD currently. Take a look at section 6:00 of the video and see for yourself the difference in continued target acquisition of SMD (green box) in Low vs Med & High especially when comparing to video as seen in 11:36. Now with highly defined targets, close to camera this makes little difference but for continued identification across further distance or as a target moves further away this becomes critically important to ensuring you continuously identify that target. Hope that made sense.
  • IVS (for those that saw issues) is working fine in this release (please see the IVS section for tripwire testing) and responsiveness to target has been improved
  • SmartIR there were barely any changes on this FW. Due to this, although SmartIR was much improved in 8/14, as I noted this primarily impacts face on installations which for most people may not give them what they need as generally installing these slightly higher around 6.5 to 7.5 ft. Therefore have again raised this to Dahua to fix IR response and strength for those situations where the target may be center of camera but at an angle other than face on

Therefore recommendation for now would be to use this as its the best FW so far BUT and I do stress BUT, there is still work needed here to get this working as expected and for that both Dahua & Andy are engaged, THANK YOU to them. I will look at the next FW once made available in September.


Video Timestamps




  • 00:28 - SmartIR Testing
  • 06:00 - SMD & IVS Testing
  • 11:28 - SMD Only Testing
  • 12:46 - SMD & IVS Full Color Test



HTH, let me know with any questions as always

I’m sure Wildcat_1 you now have a strong following of interest in your testing, perhaps like me waiting for it to conclude before buying more 5442s. My personal view is that Dahua should first nail the problem of facial white clipping when using IR without supplementary lighting - when recording 24/7 the most crucial thing is identification.

Dynamic control of the IR power appears to be virtually non existent and very limited in respect of the shutter&gain. It would appear that sensing of white clipping in real time is tricky although using it to control independent mechanisms (shutter, gain and IR power) must be done with care to avoid hunting (each one responding in the opposite sense caused by the other). In this application, however, there is the opportunity to separate the control time constants to avoid that problem.

To take an alternative view, if a bright light was switched on somewhere in the scene, we would be the first to criticise the fact that the required image of the person then became too dark.

Perhaps on reflection, the range of backlight settings that already exist are sufficient.
 
  • Like
Reactions: djernie
Hey guys, question I been using my camera to determine mounting positions and lens, My question is when you zooming this camera in and out from 2.7mm to 12mm the zoom range shows 0 to 2275 what exactly is this number indicative of? My math say a 50% zoom or 1137.5 on that scale should be about 7mm. I came up with 7, because 0 actually starts at 2.7mm for a full zoom difference of 9 mm, I know this is not the ideal way but its the programmer way in me. Either way I am just wondering what the number (0-2275) actually represents. Most of my cameras are going to be approx 2.7-3mm. I think I only flagged 2 at needing to be either 4 or 6 mm and maybe one more varifocal.

Couple of new questions: how significant is the difference between the turret version and the bullet version? I think I read somewhere the sensor was bigger on the turret. Is there a bullet version of the 54442 fixed lens?
 
  • Like
Reactions: djernie
See Focal Length Calculator

Yes, there are fixed lens versions of the bullet 5442's. The sensor on the 5442's are all 1/1.8".
 
  • Like
Reactions: djernie
Hey guys, question I been using my camera to determine mounting positions and lens, My question is when you zooming this camera in and out from 2.7mm to 12mm the zoom range shows 0 to 2275 what exactly is this number indicative of?
See: the wiki section "Determining fixed focal length requirements using a Varifocal camera on-site".
 
lol Now I understand that damn calculator last time I viewed it i misunderstood it make so much more since now that I have a camera on hand.
 
fyi my math was correct with the calculator.
 
  • Like
Reactions: samplenhold
It’s not as bad as it was. It does detect people well BUT still has issues with animals and detects them as human, dahua said there has another baseline in early Sep, we will use this one firstly
View attachment 69376
Can lower the sensitivity
View attachment 69377
Ok, so I had to reset the cam to factory defaults an additional two times. Now SMD is working, not perfectly, but much better.
 
All, I went out in the rain last night to test the latest firmware and have included a 14.5 min video below. Timestamps for ease of viewing are below.

Long story short:

Findings
  • SMD in my testing both day & night has been ‘improved’ since 8/14 BUTas @looney2ns and I agree, still has issues in identifying incorrect targets such as animals as humans (noted this to @EMPIRETECANDY last night). Therefore SMD still needs work so have sent feedback and video to Dahua in hopes to get this included in the September security baseline they are working on.
    • With regards to SMD sensitivity, there are some important takeaways here to dialing in your SMD targets for performance even while awaiting the next changes. I'll try not to lose people here as I explain what I've seen with SMD across these cameras. Specifically, sensitivity is usually understood as the higher the sensitivity, the more sensitive a given area is to an input or in this case a target BUT in the SMD implementation it appears that Dahua has taken Sensitivity as you would expect it to be implemented and combined this with 'threshold'. Take a look at regular Motion Detection settings and you will see it clearly has both Sensitivity & Threshold which are critical to any form of motion based capture. However, In SMD (although this is an effective add-on to regular Motion Detection) it is the threshold that it would appear the algorithm is using and biasing to spot targets rather than true sensitivity (as described). This means to enable more accurate captures or response from SMD we have to adjust our understanding and expectations of it a little. Specifically this means flipping the understanding so that setting to low 'sensitivity' (actually threshold) means it will trigger / find what the cam see's as targets more frequently than high sensitivity (again actually threshold) which would result in less target acquisition or a more conservative approach to identifying then tracking a target through FOV. I saw this in my testing and as I felt this was important, wanted to mention it here as it does change the way you approach SMD currently. Take a look at section 6:00 of the video and see for yourself the difference in continued target acquisition of SMD (green box) in Low vs Med & High especially when comparing to video as seen in 11:36. Now with highly defined targets, close to camera this makes little difference but for continued identification across further distance or as a target moves further away this becomes critically important to ensuring you continuously identify that target. Hope that made sense.
  • IVS (for those that saw issues) is working fine in this release (please see the IVS section for tripwire testing) and responsiveness to target has been improved
  • SmartIR there were barely any changes on this FW. Due to this, although SmartIR was much improved in 8/14, as I noted this primarily impacts face on installations which for most people may not give them what they need as generally installing these slightly higher around 6.5 to 7.5 ft. Therefore have again raised this to Dahua to fix IR response and strength for those situations where the target may be center of camera but at an angle other than face on

Therefore recommendation for now would be to use this as its the best FW so far BUT and I do stress BUT, there is still work needed here to get this working as expected and for that both Dahua & Andy are engaged, THANK YOU to them. I will look at the next FW once made available in September.


Video Timestamps



  • 00:28 - SmartIR Testing
  • 06:00 - SMD & IVS Testing
  • 11:28 - SMD Only Testing
  • 12:46 - SMD & IVS Full Color Test



HTH, let me know with any questions as always

Great job as usual man. IMO, there is little to no difference with the smartIR from the previous firmware version.
 

Attachments

  • DriveWay 2020-08-26 10.14.55.534 PM.jpg
    DriveWay 2020-08-26 10.14.55.534 PM.jpg
    310 KB · Views: 66
This is 5442 varifocal in my back yard with limited light color mode and no IR. The patio/hottub light was on but that is behind the camera and the light is mostly blocked by the terraces/vines around the tub. Have not really play yet with the IR features because I been following this thread about the firmware needing some work with it. So was waiting to see how the recent version shook out. I did not make any tweaks for lighting. This is the factor defaults settings other then the night/day mode.
20200816_204456.png
 
@Dave Lonsdale i can categorically state there has only been minor improvements in SmartIR specifically in camera response to fixed changes that Dahua made. In other words if you compare my 8/25 vs 8/14 super close you will see the camera adjusts quicker to objects in 8/25. However the strength of IR to objects center of image but at an angle (angle is key here in how the algorithm is tunes) still needs work. Otherwise you end up with the IR washout that is still seen today.

SMD however has improved but as I mentioned I think it can be tweaked further hence ensuring the feedback, videos and reference to line level code on the algorithms used, I’m assuring is shared with Dahua engineers in hopes we see continued improvements here.

HTH
 
  • Like
Reactions: looney2ns
Ok so I have been reading up some, does install an SD card improve SMD or is that only if your not using an NVR or BI with direct write? I am using BI with direct write, and was also considered adding deep learning. I am currently on the 8/14 firmware, I like to wait abit before doing firmware updates, but I did note some issues/delay with the motion triggering while working out my angle and lens needs with the 8/14 firmware. I just have not taken the time to sort out optimizing the SMD stuff yet.
 
An SD card is backup storage. As you can log into the cam with it's web interface and view what is stored on the SD card. You either need an SD card for the SMD functions to record to, or a compatible Dahua NVR to use the benefit of the SMD.

Ok so I have been reading up some, does install an SD card improve SMD or is that only if your not using an NVR or BI with direct write? I am using BI with direct write, and was also considered adding deep learning. I am currently on the 8/14 firmware, I like to wait abit before doing firmware updates, but I did note some issues/delay with the motion triggering while working out my angle and lens needs with the 8/14 firmware. I just have not taken the time to sort out optimizing the SMD stuff yet.