All, as promised, I am detailing the 10 bugs I’ve found with the current 5x42 / B/t5xIR S3 cameras below. In some cases the bug is in how the camera interacts with NVR’s looking to read its capabilities and accurately interpret the metadata from the camera. In each case, I’ve listed the issue I found, any conclusions I can make along with sharing in depth config and code findings with the Dev Teams + my recommended fixes.
Note: that some of these have already been addressed, some are actively being addressed and others are limitations that can't or won't be addressed.
Bug #1 - ANPR Tagging / OCR of Plates is Only Partially Implemented on the S3 Cameras Today As is the NVR Processing of S3 Fed Metadata
Detail
Bug #2 - S3 Varifocal Cameras are Missing the Manual Focus control in the Camera Menu. Specifically the Focus & Zoom option is missing completely.
Detail
Bug #3 - Self Learning 'Experience Database' in 5x42 S3 Camera IVS Menu Displays a 'Value Cannot Be Empty' Error
Detail
Bug #4 - Too much sensor / image noise on 5x42 S3 cameras
Detail
Bug #5 - Turning OFF AcuPick mode causes the camera settings (Image/Exposure/other Settings) to be reset to Default
Detail
Bug #6 - In Black & White mode at night, any exposure setting above 1/250 features banding/flickering in the image
Detail
Bug #7 - S3 Cameras Support AcuPIck 2.0 BUT AI Rated NVR's Like 5216-XI / 608-XI Currently Only Feature AcuPick 1.0
Detail
Bug #8 - No IVS Parameter Filtering or Configuration for ANIMALS on Dahua NVR's BUT S3 Cameras DO Support this Target Type
Detail
Bug #9 + #10 - AcuPick 2.0 Can Incorrectly Identify Stationary Vehicles as Active Targets & Tree's or Vertical Objects as Humans on S3 Cameras
Detail
That's the list for now. My thanks as always to the Dahua Dev Teams that have been working with me as like I said, some of these are getting improvements, fixes etc as the FW matures for the S3 + NVR devices associated with these bugs. In other cases, longer term fixes from my recommendations are being reviewed with possibility of implementation where possible by current HW constraints etc. Also thanks to @EMPIRETECANDY as both of us continue to help drive and prioritize these as I find them across my reviews, in depth tests and deployments.
While it may seem like a long list, trust me this is nothing in comparison to what I've built, identified fixes for and worked on for other releases of surveillance HW in this forum by the manufacturers in question.
Long story short, the S3 cameras really are a great all round solution that I've been using them in many deployments and pushing to their limits. With a few tweaks to FW on current versions + recommendations to improve even further in an S4 variant, these are turning into a 1 stop option (with associated models) for a number of people and their deployments.
As always, hope this was helpful.
WC
Note: that some of these have already been addressed, some are actively being addressed and others are limitations that can't or won't be addressed.
Bug #1 - ANPR Tagging / OCR of Plates is Only Partially Implemented on the S3 Cameras Today As is the NVR Processing of S3 Fed Metadata
Detail
- ANPR IS supported by the S3 cameras and recognized as doing so by Dahua NVR's (EI/XI etc) when using either Video Metadata and AcuPick modes on these cameras.
- However there are 2 problems:
- When reviewing Video Metadata or AcuPick captures on the camera (using Picture menu) you will see vehicle caps + ANPR plate cutout captures, however there is nothing captured in the Record (video) menu at all.
- Similarly on the supported NVR’s, the plate capture is clearly seen (and highlighted in a Plate cutout on the AI Search screen) but shows UNKNOWN for Plate No. and has NO vehicle details even when the plate (and vehicle) is clearly legible. You have to enable the AI NVR (XI for example) to do Video Metadata by Recorder to get around this issue which a) is not using the metadata from the camera (separate metadata and captures) and b) utilizes an AI processing stream on NVR for no good reason.
- To End Users - Enable Video Metadata by Recorder on supported NVR's
- To Dahua - Ensure that the Metadata that is tagged in the capture files (I've shown it's present at point of capture to devs) a) gets correctly set as an ANPR capture in the Record (Video) menu + update the NVR FW to correctly read and tag the associated metadata from the camera.
Bug #2 - S3 Varifocal Cameras are Missing the Manual Focus control in the Camera Menu. Specifically the Focus & Zoom option is missing completely.
Detail
- Focus & Zoom menu IS included on Z4 model but not on any other varifocal model. It should be because there are many situations where you want to remove auto control from the camera and instead set your own Focus Mode
- To Dahua - Enable this menu across the varifocal range of cameras. It does exist but hidden in code except for Z4 etc.
Bug #3 - Self Learning 'Experience Database' in 5x42 S3 Camera IVS Menu Displays a 'Value Cannot Be Empty' Error
Detail
- The error means that for some unknown reason, the video you are attempting to have the camera learn from is not able to be processed
Bug #4 - Too much sensor / image noise on 5x42 S3 cameras
Detail
- I traced this back to Dahua baseline configuration and backend sensor adjustments. Recommended fixes here in code + config.
- To Dahua - I provided options in code + config to adjust and build into baseline.
Bug #5 - Turning OFF AcuPick mode causes the camera settings (Image/Exposure/other Settings) to be reset to Default
Detail
- Previous settings are lost when switching between these 2 modes on early FW.
- To Dahua - Identified where this was seen and recommended way to keep persistent. The AcuPick (vs other) modes operates on different branches within code therefore my recommendation is also to ensure these are merged in specific areas while keeping the mode specific functionality separate
Bug #6 - In Black & White mode at night, any exposure setting above 1/250 features banding/flickering in the image
Detail
- You’ll notice flickering on your image when you use any exposure above 1/250 at night and any light is present either directly or off screen. As this is present in high shutter speeds this is not simply exposure differences between bright and dark areas but rather issues with the shutter and exposure config possible from the camera HW.
- Rolling shutter plays a huge part here and implementation of Global Shutter / Exposure would assist.
- This is a limitation of the camera
Bug #7 - S3 Cameras Support AcuPIck 2.0 BUT AI Rated NVR's Like 5216-XI / 608-XI Currently Only Feature AcuPick 1.0
Detail
- Feature disparity between S3 cameras and current AI NVR's.
- S3 cams feature AcuPick 2.0 support BUT 5216 / 608 XI NVR’s are limited (in current FW) to 1.0 therefore not allowing the cams to be used to their potential.
- I have been working on beta FW tests for new NVR FW to address this, found some additional bugs there BUT long story short, this WILL be implemented soon.
- I will do a post on the new NVR beta FW, my thoughts on that + bug list.
Bug #8 - No IVS Parameter Filtering or Configuration for ANIMALS on Dahua NVR's BUT S3 Cameras DO Support this Target Type
Detail
- Feature disparity with the S3 camera being able to accurately identify animals as part of its IVS target filtering BUT the NVR’s NOT being able to filter by this target type. This reduces the value of this target capability by camera as the NVR you attach to cannot use or display this tag.
- I was able to trace this back to the fact that NVR’s do NOT support perimeter post intelligence therefore only People, Vehicles and Non Motor Vehicles are currently supported through IVS.
- Therefore at this time, Animals will only be a tagged target type via camera and not the NVR.
Bug #9 + #10 - AcuPick 2.0 Can Incorrectly Identify Stationary Vehicles as Active Targets & Tree's or Vertical Objects as Humans on S3 Cameras
Detail
- While incorrect target capture is expected, this is why the Self Learning 'Experience Database' was introduced (and is really good BTW), to improve target acquisition accuracy.
- In my testing, I’ve seen 100’s of captures in the space of 1 hour that are incorrect with AcuPick unfortunately.
- However in further testing across scenarios it appears these cameras are purposefully limited to Self Learning from IVS based captures and NOT AcuPick identified clips due to the limited processing power.
- So in summary I can confirm that IVS = Self Learning (through Experience Database), AcuPick = cannot self learn.
- For the record, AcuPick Self Learning is not even supported on the more powerful Wizmind X cameras.
- To Dahua - I've recommend that with AcuPick becoming a standard across the portfolio that this is something that Dahua considers as part of future HW considerations in design.
- To Dahua - In both cases for Bug #9 and Bug #10 I’ve provided Dahua with a number of code level change recommendations for the VehicleTrait and HumanTrait AI algorithms that could be used to optimize this algorithm and reduce the incorrect target acquisition.
That's the list for now. My thanks as always to the Dahua Dev Teams that have been working with me as like I said, some of these are getting improvements, fixes etc as the FW matures for the S3 + NVR devices associated with these bugs. In other cases, longer term fixes from my recommendations are being reviewed with possibility of implementation where possible by current HW constraints etc. Also thanks to @EMPIRETECANDY as both of us continue to help drive and prioritize these as I find them across my reviews, in depth tests and deployments.
While it may seem like a long list, trust me this is nothing in comparison to what I've built, identified fixes for and worked on for other releases of surveillance HW in this forum by the manufacturers in question.
Long story short, the S3 cameras really are a great all round solution that I've been using them in many deployments and pushing to their limits. With a few tweaks to FW on current versions + recommendations to improve even further in an S4 variant, these are turning into a 1 stop option (with associated models) for a number of people and their deployments.
As always, hope this was helpful.
WC
...