Dahua IPC-HDW5442T-ZE 4MP Varifocal Turret - Night Perfomance testing
- Thread starter Wildcat_1
- Start date
biggen
Known around here
- May 6, 2018
- 2,693
- 3,067
tdwilli1
Getting the hang of it
- Aug 25, 2016
- 111
- 51
biggen
Known around here
- May 6, 2018
- 2,693
- 3,067
Still under testing, when test out, will release it here.
I just wanted to update that I tested the latest 'release' FW and it is not good from an IR perspective. It appears that none of the IR updates we tested and pushed in the test release made it into this release version therefore IR washout is terrible. I have made sure this feedback is passed to @EMPIRETECANDY and Dahua. In the meantime however, the AI algorithm's were updated in this release version so I will continue to test SMD today as that was the other issue that I and @looney2ns reported which specifically was SMD capturing leaves, cats, lights instead of filtering those out correctly.
So in summary, I will report back on the SMD performance in this 'release' version but for now, my recommendation is DO NOT USE DH_IPC-HX5XXX-Volt_MultiLang_NP_Stream3_V2.800.0000000.20.R.200903. Instead use the last test version (with improved IR and based on 5/29 SMD algorithm), this version is General_IPC-HX5XXX-Volt_Eng_N_Stream3_V2.800.15OG002.0.R.200903. If I find that SMD is improved in the release version I have already made sure @EMPIRETECANDY and Dahua are aware of current issues with this release and are ready to take the baseline of 9/03 (release) for SMD and the IR adjustments from 15OG002 (test version) and combine the 2 into a complete release. Its a real shame that it wasn't merged into this release version already with the work we've put in BUT Dahua are aware (THANK YOU) so hope we will see something soon.
So in summary, I will report back on the SMD performance in this 'release' version but for now, my recommendation is DO NOT USE DH_IPC-HX5XXX-Volt_MultiLang_NP_Stream3_V2.800.0000000.20.R.200903. Instead use the last test version (with improved IR and based on 5/29 SMD algorithm), this version is General_IPC-HX5XXX-Volt_Eng_N_Stream3_V2.800.15OG002.0.R.200903. If I find that SMD is improved in the release version I have already made sure @EMPIRETECANDY and Dahua are aware of current issues with this release and are ready to take the baseline of 9/03 (release) for SMD and the IR adjustments from 15OG002 (test version) and combine the 2 into a complete release. Its a real shame that it wasn't merged into this release version already with the work we've put in BUT Dahua are aware (THANK YOU) so hope we will see something soon.
Looks like i will make Andy's firmware for my own cams.
Thanks @Wildcat_1 Will spend more time on the firmware and we work out it. I will delete the DH one.
Thanks @Wildcat_1 Will spend more time on the firmware and we work out it. I will delete the DH one.
@EMPIRETECANDY , if you are going to delete the DH one, can you point us to the next best version?
Last edited:
All, I wanted to provide an update on my testing of the latest 'release' version as I tested against SMD today.
I'm pleased to say that in my testing the SMD works better in this release and results in much fewer false alerts/triggers. There is one situation that still freaks it out which is if you have a foreground object (tree in this example) that is blowing in the wind and a background object behind it (parked car) then as the branches/leaves move and clear the FOV to reveal the car, the SMD triggers even though its stationary. In other words it see's this as a target. While not a deal breaker by any means it can lead to multiple caps on a windy day. Therefore would be great to see Dahua implement a similar feature as they did to People Recognition with the 'living' AI algorithm. Maybe for cars it can be stationary vs moving within x. Either way I've provided this feedback and some details on how to potentially implement as well in efforts to continue to improve the AI base algorithm.
So, in summary, the latest release is terrible for IR, better for SMD (best release for SMD in my testing so far). Therefore I still do not recommend that you install the release version but instead the last test version @EMPIRETECANDY posted.
In the meantime I've fed back to Andy & Dahua that my recommendation based on what I see in code and performance is to build the next FW (say October for example) using IR algorithm and adjustments from the last test FW (15OG002.0 for those wanting details), tweak strength down slightly still for off angle targets center of cam and then combine the SMD AI baseline algorithms from the latest release FW (20.R.200903) with tweaks (if possible) to counter the stationary car/moving foreground object and other related scenarios. If that combination utilizing both FW components I list (along with tweaks) is followed then I believe the next release should be on point. Will report back when I hear feedback and will test again when next test is available.
HTH
I'm pleased to say that in my testing the SMD works better in this release and results in much fewer false alerts/triggers. There is one situation that still freaks it out which is if you have a foreground object (tree in this example) that is blowing in the wind and a background object behind it (parked car) then as the branches/leaves move and clear the FOV to reveal the car, the SMD triggers even though its stationary. In other words it see's this as a target. While not a deal breaker by any means it can lead to multiple caps on a windy day. Therefore would be great to see Dahua implement a similar feature as they did to People Recognition with the 'living' AI algorithm. Maybe for cars it can be stationary vs moving within x. Either way I've provided this feedback and some details on how to potentially implement as well in efforts to continue to improve the AI base algorithm.
So, in summary, the latest release is terrible for IR, better for SMD (best release for SMD in my testing so far). Therefore I still do not recommend that you install the release version but instead the last test version @EMPIRETECANDY posted.
In the meantime I've fed back to Andy & Dahua that my recommendation based on what I see in code and performance is to build the next FW (say October for example) using IR algorithm and adjustments from the last test FW (15OG002.0 for those wanting details), tweak strength down slightly still for off angle targets center of cam and then combine the SMD AI baseline algorithms from the latest release FW (20.R.200903) with tweaks (if possible) to counter the stationary car/moving foreground object and other related scenarios. If that combination utilizing both FW components I list (along with tweaks) is followed then I believe the next release should be on point. Will report back when I hear feedback and will test again when next test is available.
HTH
I'm losing the plot here guys. I decided to buy 5442s to get good image quality. From my standpoint, I record 24/7 and using my 5442s to identify an intruder at night when it's dark is far more important than flawless SMD. I have held off from updating my firmware and am suspicious of all recent updates. Referring to looney2ns snapshots, is it not the case that 6/20 FW is far better in respect of washout problems than 8/25, 9/03 or any other version that's kicking about?
@Dave Lonsdale
You're not losing the plot, just a lot to follow
In answer to your specific question, 6/20 wasn't bad from IR perspective but was terrible for SMD. 7/09 featured further IR tweaks but was based on 6/20 baseline hence still had bad SMD. Therefore 9/03 was an ask to bring final tweaks together for IR (some things still needed tweaking as I mentioned) and solve the terrible SMD (utilizing 5/29 as baseline and adjusting SMD to a better algorithm from there). In its current iteration as I mentioned above, the 9/03 release version is not in a good state hence my recommendations, findings and feedback to Dahua which I hope will be addressed in an October release.
Does that help clarify a little more ?
You're not losing the plot, just a lot to follow
In answer to your specific question, 6/20 wasn't bad from IR perspective but was terrible for SMD. 7/09 featured further IR tweaks but was based on 6/20 baseline hence still had bad SMD. Therefore 9/03 was an ask to bring final tweaks together for IR (some things still needed tweaking as I mentioned) and solve the terrible SMD (utilizing 5/29 as baseline and adjusting SMD to a better algorithm from there). In its current iteration as I mentioned above, the 9/03 release version is not in a good state hence my recommendations, findings and feedback to Dahua which I hope will be addressed in an October release.
Does that help clarify a little more ?
sebastiantombs
Known around here
bigredfish
Known around here
What's your recording setup like: NVR/BI? what about retention schedule and recording drive space? Just curious, as I was planning on relaying on the motion sensors to trigger recording, but if I can not get sensitive right I just might switch to continuous. Right now I am using BI, 4 cams, 6TB of recording drive space. Not overly concerned with drive space usage as I could add more if need be, my original plan was to retain 30 days but not at continuous recording.
yes, I can confirm that my dog is still staring in the SMD of the 20200903
I always thought that my dog is almost human and now AI machine confirm it. but according to looney2ns, he is not THAT special.
SORRY, CORRECTION:
There is now the general FW and the one for 5442T-ZE. I'm using both 5442T-ZE and B5442E-ZE. I was updating both models.
There's one cam that detects my dog more than others. So far, that cam was the one detecting my dog BUT that one happens to be a B5442E-ZE cam.
I can confirm though, that my 5442T-ZE detects it's favorite tree as it did before. even in a none MD marked area.
I could promise to update you if and when my dog is captured also on the T-ZE cam, but @looney2ns has already reported it does so it's pointless.