All
As promised, I’ve completed my testing of this FW (22-02-18). I tested this against 4 x 5442’s in the end (3 turrets with different HW versions + 1 bullet cam) and other VOLT cameras including the 5241-Z12. As long as you use this FW with 5442’s only (based on my findings and issues with 5241 & other VOLT cams) then in summary, I do see improvements here.
Let me break down some of the key areas below:
SmartIR - This is now more on point to the recommendations I made last year and that we saw implemented early on BUT needed a little further work based on additional feedback I sent Dahua inline with SMD releases branches. Glad to say this has now made it into this release and overall works pretty well (see quick video below). With that said, there is 1 issue I have with it and that is the transition of SmartIR isn’t as smooth as I’d like. To be clear (and you can see it in the video below) its a more pronounced step ladder in its approach to kicking up or down IR strength BUT certainly the ending result (no IR washout) is what we all want to see and a marked improvement on previous SMD based releases. I am feeding back the ‘step ladder’ IR approach to see what can be done there in future to adjust it to be smoother / quicker as it was on the early SmartIR FW I tested last year after my feedback and approved wholeheartedly. Again, this is not a dealbreaker BUT I think with a little more tweaking this can be perfect. I'm also testing a 5842 SmartIR FW and on that it works very smoothly so will be looking into implementation in that code to identify differences between these for Dahua.SMD 3.0 - I also saw improvements here in my testing. Specifically in the areas of Target Identification (where I saw no false hits, even from deer in multiple days of testing) and in Target Acquisition Speed. To be clear, thats not to say that you'll never see false hits BUT in other versions I would have typically experienced some within this longer testing period. The latter (Target Acquisition Speed) has improved albeit I still think a little more work can be done here which would assist in tighter FOV’s. Either way this release represents an overall improvement here in my testing. As an example, I was seeing vehicles tagged as SMD hits and in reviewing the clip wondered IF the algorithm would actually target correctly and sure enough even at a distance of over 100’ from cam, the target would be acquired and tracked even as it passed foreground objects. This worked well with Human identification as well. From an overall SMD 3.0 perspective on these cams, while I think a little tweaking can be done here, I want to be clear that I think this is getting close to the limits of what is possible (theoretical max If you will) with current SOC and algorithm in use on this range of cams. Will be interesting to see SMD 4.0 this year and how that performs (I feel a comparison feature video coming up SMD 3.0 vs 4.0 etc)
IVS - Similar to SMD, I saw improvements in both identification and acquisition although depending on HW version and IVS rule type, this is still a little slow. While I didn’t see any mis-categorizations on this FW so far, I DID see the speed of acquisition was still somewhat delayed (for clarity: improved but delayed) especially when pushing the camera. Specifically, I noticed this more so on Tripwire rules where in a perfect storm scenario test I did (distance to target about 80ft, target crossing right to left of FOV) that there was about a 1-2 second delay in target box acquisition. This can be more problematic with super tight FOV's as although you have a more clearly definable target, if your FOV only allows it a limited time within the FOV, then you might miss a cap. Just remember pre-record is your friend here and be careful that if you rely heavily on Trips + have a tight FOV with limited time on target, to configure your rules accordingly for improved success. Reach out to me if you need any assistance on that.
So, overall, definitely an improvement over the last FW in the areas above, just remember the caveats I call out to set expectations correctly. I’m going to feed back these additional findings to Dahua as well and have already let them know about the issues with ‘other’ VOLT based cameras such as the 5241 for further investigation by them.
HTH and as always, let me know with any questions
SmartIR Test Video - No Light On Scene Except In Background
@Wildcat_1 , thanks for your insights on this new firmware, although I sometimes get the impression that your words are very carefully chosen to limit any strong negative assessments, and potential for subsequent backlash from our favorite suppliers and camera manufacturers. For example, the broad statement that only tells us "I saw improvements in both identification and acquisition although depending on HW version and IVS rule type, this is still a little slow" leaves out the details of which hardware revisions are more negatively impacted: the newer ones, or the older/early release 5442 cameras.
Still, great to see that some progress is being made with firmware tweaks. In your opinion, is the December 2020 version still the best performing in terms of speed and accuracy of target acquisition in IVS rules, or would you characterize this new one as superior in that regard?
