Dahua Ultra Starlight Varifocal Bullet (IPC-HFW8232E-Z)

[Dragon]

Day/Night profile switching (not to be confused with color/B&W mode switching) won't activate based on light, just time. Fairly useless unless I'm missing something here..

Unfortunately, you're not missing anything. People keep asking for this feature and Dahua has not yet provided. HIKvision and even many no-name cameras have two settings based on light level, but not Dahua. Really maddening. The only way to switch settings based roughly on light level (ignoring weather) is to use an external device to send the camera a command to switch around sunset and sunrise. The process is described here. Additional methods to get sunrise/sunset time are described here.

I'm running Shutter Priority 1/2000 during the day and 1/60 at night. Nayr said somewhere he prefers Auto mode at all times, but for me, Auto mode in the day leaves cars blurry and at night it leaves moving people blurry.

 

[CamCrazy]

Unfortunately, you're not missing anything. People keep asking for this feature and Dahua has not yet provided. HIKvision and even many no-name cameras have two settings based on light level, but not Dahua. Really maddening. The only way to switch settings based roughly on light level (ignoring weather) is to use an external device to send the camera a command to switch around sunset and sunrise. The process is described here. Additional methods to get sunrise/sunset time are described here.

I'm running Shutter Priority 1/2000 during the day and 1/60 at night. Nayr said somewhere he prefers Auto mode at all times, but for me, Auto mode in the day leaves cars blurry and at night it leaves moving people blurry.

I wish Dahua would take note of this and resolve their camera settings, it is extremely frustrating and surely something they could remedy. Totally agree that leaving in auto will in anything other than good to excellent light result in blurred motion, similarly I run 1/500 to 1/1000 during the day and 1/250 or above at night depending on the camera. Having to manually schedule the profile switching is madness, we shouldn't have to use external methods to schedule for sunrise/sunset and so on.

 

[Dragon]

One odd thing I realized about this camera today is shutter priority of say 1/2000 is not a minimum, it's a fixed value. Even if light is so bright that 1/4000 or higher with a different iris size would keep light and dark more balanced, it will still use 1/2000 if you set it there. This is different from other cameras I've used. So it's best to use the "Customized Range" shutter priority setting to set a min/max speed.

I've tried 1/500 to 1/2000 during the day and only 1/2000 eliminates all blur on cars in my case, so I set min shutter to 1000/2000=0.5ms. Max shutter on the camera is 1/100000, so 1000/100000=0.01ms. It wants the smaller number first in the UI even though the smaller number represents faster shutter, so I set the range to 0.01 ~ 0.5 during the day.

For me, at night, 1/250 produces sharp figures but with a lot of jagged edges and other artifacts. If you turn off 3DNR you can see the image is very grainy, so all the artifacting is caused by 3DNR smoothing the grains into weird shapes. With the amount of light I have to work with (really just IR, no street lights), 1/60 to 1/120 (8.33 ~ 16.67 shutter range) seems the best compromise to minimize artifacts without creating blur on people. Cars are hopeless at night. I don't allow higher shutter at night because then a light shone into the camera could set a high shutter that makes everything else black. I haven't tested if that's an actual problem but that's my theory for 1/120 max shutter (EDIT: It's a bad idea. See later post).

 

[CamCrazy]

One odd thing I realized about this camera today is shutter priority of say 1/2000 is not a minimum, it's a fixed value. Even if light is so bright that 1/4000 or higher with a different iris size would keep light and dark more balanced, it will still use 1/2000 if you set it there. This is different from other cameras I've used. So it's best to use the "Customized Range" shutter priority setting to set a min/max speed.

I've tried 1/500 to 1/2000 during the day and only 1/2000 eliminates all blur on cars in my case, so I set min shutter to 1000/2000=0.5ms. Max shutter on the camera is 1/100000, so 1000/100000=0.01ms. It wants the smaller number first in the UI even though the smaller number represents faster shutter, so I set the range to 0.01 ~ 0.5 during the day.

For me, at night, 1/250 produces sharp figures but with a lot of jagged edges and other artifacts. If you turn off 3DNR you can see the image is very grainy, so all the artifacting is caused by 3DNR smoothing the grains into weird shapes. With the amount of light I have to work with (really just IR, no street lights), 1/60 to 1/120 (8.33 ~ 16.67 shutter range) seems the best compromise to minimize artifacts without creating blur on people. Cars are hopeless at night. I don't allow higher shutter at night because then a light shone into the camera could set a high shutter that makes everything else black. I haven't tested if that's an actual problem but that's my theory for 1/120 max shutter.

Dragon, thanks for the detailed information, I think the shutter range is a safer bet from what I've seen so far in general with the Dahua range, interesting you run at 1/2000 to freeze motion, do you also run that same shutter speed on your other Dahua cameras, I have found 1/500 to 1/1000 to be enough in fairly poor light on the 5231 and SD49225T-HN, as a side note I do have sharpness off (set to 0) and 3DNR is set to 10 or below, 5 gets rid of the worst from what I've seen. I have actually experimented with WDR running on my 5231's these last few days, only on value between 4-6 but this has had no apparent negative effect that I can see.

Seems that any additional sharpness or 3DNR can have a effect on what you expect quality wise from still frames. Sometimes I do find the iris does odd things on both the 5231's and SD49225T-HN in that it will appear to stay open too much until settings are changed/saved then it resorts back to what you would expect. So far I have found the SD49225T-HN by far the best for freezing motion at relatively low shutter speeds which seems a bit odd. Freezing motion with cars at night even with lots of additional IR is a tall order, especially if they are moving across the frame or diagonally, was hoping the 8232E-Z and to some degree the 8231E-Z would be an improvement in this area over the lesser models, since I am looking to order one or more of the 8232/31 cameras I'm watching this with interest!

 

[Dragon]

I found that with shutter limited to 1/60 ~ 1/120 at night, you walk up to the camera with external IR and your face becomes unrecognizable white. With the shutter set to 1/60 ~ 1/100000, the camera darkens everything to keep your face recognizable. Shining a flashlight doesn't trick it - it still keeps things pretty balanced. However, the flashlight may send it into color mode, at which point everything looks terrible. This is one advantage of using an external API call to force it into Night profile at sunset, then set the night profile to B&W only mode.

Dragon, thanks for the detailed information, I think the shutter range is a safer bet from what I've seen so far in general with the Dahua range, interesting you run at 1/2000 to freeze motion, do you also run that same shutter speed on your other Dahua cameras, I have found 1/500 to 1/1000 to be enough in fairly poor light on the 5231 and SD49225T-HN, as a side note I do have sharpness off (set to 0) and 3DNR is set to 10 or below, 5 gets rid of the worst from what I've seen. I have actually experimented with WDR running on my 5231's these last few days, only on value between 4-6 but this has had no apparent negative effect that I can see.

I don't have any other Dahuas. I have an HIKvision that does okay with lots of external IR and a no-name Chinese "p-iris" cam people were briefly excited about until I paid too much for it to find it doesn't have p-iris and is mostly terrible for the price.

I tried reducing 3DNR from 50 to 45 at night and artifacts are immediately visible. If I turn off 3DNR during the day, I see no difference. So the "great" night vision of this camera is strongly dependent on taking a very noisy image and smoothing it out with 3DNR. 3DNR itself means using data from previous frames to combine with the current frame, which keeps static portions of the image almost completely noise-free and looks far better than any of my other cams (basically the same idea as slow shutter). At the same time, it can't do that with parts of the image that move, so it uses some sort of nearest-neighbor algorithm to reduce noise in those parts. Unfortunately, that algorithm can't produce great results on a noisy image so you get more artifacts on moving objects, which are also the things you usually care most about in a security context. I haven't altered sharpness but I'm sure it's not going to have any magical effect - the core problem is the night image is very noisy and you can't extract a perfect image from that.

Freezing motion with cars at night even with lots of additional IR is a tall order, especially if they are moving across the frame or diagonally, was hoping the 8232E-Z and to some degree the 8231E-Z would be an improvement in this area over the lesser models, since I am looking to order one or more of the 8232/31 cameras I'm watching this with interest!

I can set the shutter to 1/1000 or 1/2000 at night and get cars that look unblurry, yet also contain obvious noise and jagged parts. Somehow, the license ends up a solid block of white. There is a point where the license is near the edge of IR where it's almost visible. Maybe if I got just the right settings or just the right level of IR it might make the license visible, but it would be a process. Licenses actually aren't that easy to read even during the day. The camera does this weird thing where most of the letters are clear but then it puts a "blurry block" covering some letters. The blurry block looks like jpeg/mpeg compression, yet I have the kbps set to max at 10fps and the recording software says I'm hitting like half of max bitrate so I don't know why I'm getting compression artifacts other than the camera has a poor encoder. I'm using Milestone Xprotect 2016 - I assume it doesn't re-encode anything but just writes what the camera sends it. Camera quality is set to max in Xprotect so maybe it can reencode based on that quality setting? Another thing to research I guess.

At this point, I'm rather tired of this process of gradually buying more expensive cameras only to find they still don't always make people and cars identifiable at night (or during the day, in some cases). So I'm about ready to pull the trigger on a Bosch Dinion Starlight 8000 5MP (model NBN-80052-BA) for $1200 with 4-13mm p-iris lens (model LVF-8008C-P0413). This shows a hi-res, wide angle scene where everyone is crisp and even when they turn off 80% of the lights, people are still recognizable even if some blur. What strikes me about it is how clean everything is. You see each pixel. Nothing is jagged or artifacted. When someone stops moving and looks at the camera, you see their face clearly even if it's only 20 pixels across. This video shows just how low light the latest Bosch cameras can go without blurring. The 1920x1080p cam in that video that works in the lowest light is probably Starlight 7000 (model NBN-73023-BA) or 6000 (model NBN-63023-B) which cost $750 and $576 without lens. Lens adds ~$100 to $200. You also get what looks like true object tracking and intrusion detection instead of the lame, rarely works if at all implementations found on the underpowered HIK and Dahua cams.

One thing I haven't found an example of with Bosch 8000 is IR illumination with movement. People like taking these videos with no illumination and marveling at how clear everything is, except there's a lot of movement blur so who knows how something like a car license plate would turn out with IR illumination. Even networkcameracritic failed to provide a high-shutter speed night movement video. So maybe I'm still chasing my tail with dreams of night identification, but at least I'll know if Bosch can't do it, probably nothing on the market can (except maybe that $3.5k GBO 1" sensor camera someone mentioned, but I think that truly is more than I'm willing to spend).

 

[CamCrazy]

I found that with shutter limited to 1/60 ~ 1/120 at night, you walk up to the camera with external IR and your face becomes unrecognizable white. With the shutter set to 1/60 ~ 1/100000, the camera darkens everything to keep your face recognizable. Shining a flashlight doesn't trick it - it still keeps things pretty balanced. However, the flashlight may send it into color mode, at which point everything looks terrible. This is one advantage of using an external API call to force it into Night profile at sunset, then set the night profile to B&W only mode.



I don't have any other Dahuas. I have an HIKvision that does okay with lots of external IR and a no-name Chinese "p-iris" cam people were briefly excited about until I paid too much for it to find it doesn't have p-iris and is mostly terrible for the price.

I tried reducing 3DNR from 50 to 45 at night and artifacts are immediately visible. If I turn off 3DNR during the day, I see no difference. So the "great" night vision of this camera is strongly dependent on taking a very noisy image and smoothing it out with 3DNR. 3DNR itself means using data from previous frames to combine with the current frame, which keeps static portions of the image almost completely noise-free and looks far better than any of my other cams. At the same time, it can't do that with parts of the image that move, so it uses some sort of nearest-neighbor algorithm to reduce noise in those parts. Unfortunately, that algorithm can't produce great results on a noisy image so you get more artifacts on moving objects, which are also the things you usually care most about in a security context. I haven't altered sharpness but I'm sure it's not going to have any magical effect - the core problem is the night image is very noisy and you can't extract a perfect image from that.



I can set the shutter to 1/1000 or 1/2000 at night and get cars that look unblurry, yet also contain obvious noise and jagged parts. Somehow, the license ends up a solid block of white. There is a point where the license is near the edge of IR where it's almost visible. Maybe if I got just the right settings or just the right level of IR it might make the license visible, but it would be a process. Licenses actually aren't that easy to read even during the day. The camera does this weird thing where most of the letters are clear but then it puts a "blurry block" covering some letters. The blurry block looks like jpeg/mpeg compression, yet I have the kbps set to max at 10fps and the recording software says I'm hitting like half of max bitrate so I don't know why I'm getting compression artifacts other than the camera has a poor encoder. I'm using Milestone Xprotect 2016 - I assume it doesn't re-encode anything but just writes what the camera sends it. Camera quality is set to max in Xprotect so maybe it can reencode based on that quality setting? Another thing to research I guess.

At this point, I'm rather tired of this process of gradually buying more expensive cameras only to find they still don't always make people and cars identifiable at night (or during the day, in some cases). So I'm about ready to pull the trigger on a Bosch Starlight 8000 5MP (model NBN-80052-BA) for $1200 with 4-13mm p-iris lens (model LVF-8008C-P0413). This shows a hi-res, wide angle scene where everyone is crisp and even when they turn off 80% of the lights, people are still recognizable even if some blur. What strikes me about it is how clean everything is. You see each pixel. Nothing is jagged or artifacted. This video shows just how low light the latest Bosch cameras can go without blurring. The 1920x1080p cam in that video that works in the lowest light is probably Starlight 7000 (model NBN-73023-BA) or 6000 (model NBN-63023-B) which cost $750 and $576 without lens. Lens adds ~$100 to $200. You also get what looks like true object tracking and intrusion detection instead of the lame, rarely works if at all implementations found on the underpowered HIK and Dahua cams.

Dragon, thanks for the detailed reply. In my experience both 3DNR and sharpness do not help clarity when freezing motion, I would certainly recommend you try sharpness to zero and see how that works out. Interesting about your findings with 3DNR at night when close to the camera, I will experiment with this myself since my 5231 models bleach a bit when close to them. I run all my current IP cameras at CBR and either 12 or 15 fps, bit rate 8192.

Regarding the Bosch cameras, I have been trying to establish what wide angle megapixel lens will work well on the Dahua Box style cameras in the hope that with a f1.2 lens they might come close to what I've experienced with similar analogue box cameras running f1.0 lenses, either that or like yourself I will end up trying a Bosch model with good quality lens. At this point I am pretty sure you need to be spending $900+ to do any real damage at night and expect clean images without lots of artificial light. Regarding tracking/motion/intrusion I have so far had great results with the IVS feature on Dahua cameras, motion detection has always been pretty solid but the IVS tripwires and intrusion detection controlling a PTZ is quite impressive.

 

[Dragon]

have so far had great results with the IVS feature on Dahua cameras, motion detection has always been pretty solid but the IVS tripwires and intrusion detection controlling a PTZ is quite impressive.

Okay, I shouldn't slag Dahua because I haven't yet tried their intrusion detection. I've only tried HIK's a couple years ago. The cameras are looking at an area with moving plants and shadows so it's a tough environment - maybe those with more static scenes have no trouble.

 

[CamCrazy]

Okay, I shouldn't slag Dahua because I haven't yet tried their intrusion detection. I've only tried HIK's a couple years ago. The cameras are looking at an area with moving plants and shadows so it's a tough environment - maybe those with more static scenes have no trouble.

The motion detection in areas where you have moving plants and shadows is indeed challenging, I have similar situation in certain areas and really there isn't a magic solution for this, you can control the minimum and maximum object size with IVS features but even that isn't immune to shadow and light changes. Motion in my experience with Dahua DVR's and NVR units has been solid, yes they will miss something on the odd occasion for no apparent reason but rarely. Angle and height of the camera in question will influence the success, if you are low down then the moving plants will be harder to eliminate, higher up and this becomes less of an issue potentially. So long as you have an area in the scene that is path, pavement or driveway that doesn't have moving plants hopefully you will be able to single this out for motion detection/IVS and at worst have shadows only to blame for missing motion or false alarms

 

[Dragon]

I tried setting up a line crossing detection on the street with sizes set for cars. It seems maybe 80% accurate based on the log in the camera but it's hard to tell because I can't get it to work with Milestone to review the video clips. The camera publishes its own set of events that Milestone can watch. There isn't one obviously for line crossing or the other special detection modes, so I watch the built in motion detection event hoping that's triggered on line crossing as well. Doing that, I get triggers on every movement detected in the camera.

Since I'm not sure if it's actually triggering that event on line crossing or just on general movement, I tried disabling the general movement detection with the main "Enable" button but it has no effect. I tried setting the schedule to turn off movement at all times except for a minute at midnight, no effect. I tried going to the define movement detection screen and disabling detection on all squares, and later on all squares but one in the corner, seemingly no effect. Even after clicking Save and the "Save successful" message, rebooting resets movement detection and shows everything enabled again.

So, that's the last straw. I'm sick of buggy, janky firmware even on the most expensive cameras in the most famous Chinese camera lines. I've ordered the Bosch.

Also, today was much more sunny than previous days and a big tree shadow has made most of my image too dark while the hill in the background is bright. I tried correcting with SSA and it looks good on the still image but introduces a lot of blur and weird outlining on faces in moving cars, plus makes license plates unreadable blurs. So the much-admired SSA is yet another still-frame-only or slow-movement-only marketing feature even during the day with 1/2000 shutter speed. We've seen previous examples that it's even more useless at night. I'm giving BLC a try which helps the shaded areas but pushes light areas into overexposure. As lighting changes throughout the day I suspect BLC mode will hurt the image.

 

[CamCrazy]

I tried setting up a line crossing detection on the street with sizes set for cars. It seems maybe 80% accurate based on the log in the camera but it's hard to tell because I can't get it to work with Milestone to review the video clips. The camera publishes its own set of events that Milestone can watch, so I watch the built in motion detection event, and I get triggers on every movement in the camera. I'm not sure if it's actually triggering that event on line crossing or just on general movement. I've tried disabling the general movement detection with the main "Enable" button but it has no effect. I tried setting the schedule to turn off movement at all times except for a minute at midnight, no effect. I tried going to the define movement detection screen and disabling detection on all squares, and on all squares but one in the corner, seemingly no effect. Even after clicking Save and the "Save successful" message, rebooting resets movement detection and shows everything enabled again.

So, that's the last straw. I'm sick of buggy, janky firmware even on the most expensive cameras in the most famous Chinese camera lines. I've ordered the Bosch.

Also, today was much more sunny than previous days and a big tree shadow has made most of my image too dark while the hill in the background is bright. I tried correcting with SSA and it looks good on the still image but introduces a lot of blur and weird outlining on faces in moving cars, plus makes license plates unreadable blurs. So the much-admired SSA is yet another still-frame-only or slow-movement-only marketing feature even during the day with 1/2000 shutter speed. We've seen previous examples that it's even more useless at night. I'm giving BLC a try which helps the shaded areas but pushes light areas into overexposure. As lighting changes throughout the day I suspect BLC mode will hurt the image.

Interesting info Dragon, this almost sounds like Milestone and not the camera unless the 8232 has a specific problem with firmware but pure speculation, I have no personal experience with Milestone. One thing I can tell you is for any camera to give a balanced image with bright light in one area and shadows in another will be an extremely tall order regardless of cost, BLC is good for a scene which stays the same throughout the day where you need to bring out extreme shadows and HLC useful for knocking back car lights/license plate reflection from IR at night, otherwise for normal outdoor scenes I wouldn't use them. Very interested to know how you get on with the Bosch, please feedback your findings once it arrives and I hope it lives up to expectations, certainly should do from what I've seen. I am still patiently awaiting feedback from Dahua and Computar on lenses for the box camera version of the 8232, otherwise before long I may be joining you with either a high end Bosch or Samsung equivalent box camera to replace my super low light analogue versions!

 

[Dragon]

Interesting info Dragon, this almost sounds like Milestone and not the camera unless the 8232 has a specific problem with firmware ... I am still patiently awaiting feedback from Dahua and Computar on lenses for the box camera version of the 8232, otherwise before long I may be joining you with either a high end Bosch or Samsung equivalent box camera to replace my super low light analogue versions!

I think you're right the problem is related to Milestone, but it would happen with any recording software. I had Milestone stop subscribing to the camera's motion detection event and after that, disabling motion detection in the camera worked and remained disabled after reboot. This implies the only motion event the camera offers is triggered ONLY on its standard motion detection. The problem is, the camera doesn't offer any event for line crossing detection, face detection, or any of its other special features. That basically makes the features useless unless you want to record to the camera's SD card and go use the camera's UI to review video.

Maybe if you purchased Dahua's NVR you'd have access to the line crossing events via some non-standard API, and maybe that's why they don't publish the events for standard recording software. You might be able to find a complicated workaround, maybe have the line-crossing event trigger alarm-out, but then you need a physical wire from alarm out to your recording hardware that links to your recording software. But then if you want a separate event for intrusion detection it may just be impossible. Maybe you could have it email that event and have your recording software watch the email box with some custom script. With Milestone, custom scripts require the much more expensive version of the software. No matter what, it's a huge pain and it could have been avoided by Dahua publishing an event for each of the special detection modes.

Anyway, I know all cameras have problems with backlighting and most handheld photography cams resolve it with two exposures that are merged together (basically, SSA or HDR or WDR) which creates blur if objects are moving. It does bother me that companies and individuals tend to show still scenes with things like SSA on and say "look how great this security camera is!" or they show night scenes with slow shutter and say the same thing. But these are _security_ cameras and features that only make still images look good are nearly useless.

Good luck with the lens hunt on those two cams but I'd be pretty dubious if they claimed an expensive lens would dramatically improve the camera's night abilities. The stock 8232 bullet I'm using has f1.53 max aperture, BTW.

 

[CamCrazy]

Maybe if you purchased Dahua's NVR you'd have access to the line crossing events via some non-standard API, and maybe that's why they don't publish the events for standard recording software. You might be able to find a complicated workaround, maybe have the line-crossing event trigger alarm-out, but then you need a physical wire from alarm out to your recording hardware that links to your recording software. But then if you want a separate event for intrusion detection it may just be impossible. Maybe you could have it email that event and have your recording software watch the email box with some custom script. With Milestone, custom scripts require the much more expensive version of the software. No matter what, it's a huge pain and it could have been avoided by Dahua publishing an event for each of the special detection modes.

The key here does look to be having a Dahua NVR that is IVS enabled, this in turn will play well with the IVS cameras, having to jump through hoops to emulate this kind of feature is definitely not the path you would want to be taking.

Anyway, I know all cameras have problems with backlighting and most handheld photography cams resolve it with two exposures that are merged together (basically, SSA or HDR or WDR) which creates blur if objects are moving. It does bother me that companies and individuals tend to show still scenes with things like SSA on and say "look how great this security camera is!" or they show night scenes with slow shutter and say the same thing. But these are _security_ cameras and features that only make still images look good are nearly useless.

Still cameras can work well with HDR feature, I guess it isn't impossible for this to be implemented in CCTV but would require three streams of different exposures dedicated to HDR for it to work well. Most of these current 'enhancements' do seem to have a negative effect on motion and in turn still image grabs. I agree 100% that there should be more focus on ensuring that these cameras will stop motion in most reasonable situations, sadly out of the box almost all will lean towards slow shutter speed giving what looks like a great image both night and day only to provide motion blur when a real event takes place, this is disappointing.

Good luck with the lens hunt on those two cams but I'd be pretty dubious if they claimed an expensive lens would dramatically improve the camera's night abilities. The stock 8232 bullet I'm using has f1.53 max aperture, BTW.

Thanks, I should be able to get something out of Computar hopefully, the IPC-HF8232F has an aperture of f1.2 so if that can be matched with a similar f1.2 lens which I expect to be around $100-$150 dollars equivalent it should do a better job in low light than the f1.4 and f1.53 apertures on the bullet cameras, all in theory/on paper of course !

 

[Dragon]

Hey CamCrazy, you actually had a good point about sharpness. At night, I tried reducing it to 0 and kept wondering why the picture gets so blurry. Then I realized if I turn off 3DNR, the picture gets much sharper again with some noise, but not excessive amounts. If I leave sharpness at default 50 with no 3DNR then the whole image looks like a mess of harsh noise because it's sharpening fairly subtle noise into high contrast noise.

It seems like the combined effect of 50 3DNR plus 50 sharp is sort of like putting a blur filter to get rid of all noise, then sharpening it to create an image that is smoother than the original but may lose some details in weird ways. With no sharp and no 3DNR I can actually see the blur of letters on a license at night instead of seeing solid white. Plant leaves that were otherwise solid white also have some shading. I suspect NR+sharp may have to do with the odd blocks of blurriness on license plates but have to wait till daytime to check. The night image actually seems more true with no 3DNR and 0 sharpness, though the constant low noise is probably going to trigger motion detection more and reduces stream compression effectiveness. Maybe some compromise of lower 3DNR and lower sharpness values will do better than 50 on each.

 

[CamCrazy]

Hey CamCrazy, you actually had a good point about sharpness. At night, I tried reducing it to 0 and kept wondering why the picture gets so blurry. Then I realized if I turn off 3DNR, the picture gets much sharper again with some noise, but not excessive amounts. If I leave sharpness at default 50 with no 3DNR then the whole image looks like a mess of harsh noise because it's sharpening fairly subtle noise into high contrast noise.

It seems like the combined effect of 50 3DNR plus 50 sharp is sort of like putting a blur filter to get rid of all noise, then sharpening it to create an image that is smoother than the original but may lose some details in weird ways. With no sharp and no 3DNR I can actually see the blur of letters on a license at night instead of seeing solid white. Plant leaves that were otherwise solid white also have some shading. I suspect NR+sharp may have to do with the odd blocks of blurriness on license plates but have to wait till daytime to check. The night image actually seems more true with no 3DNR and 0 sharpness, though the constant low noise is probably going to trigger motion detection more and reduces stream compression effectiveness. Maybe some compromise of lower 3DNR and lower sharpness values will do better than 50 on each.

Pleased you have seen some improvement, in testing I haven't seen an advantage having sharpness at anything more than 0-10 (mostly 0 for me) and 3DNR usually 0-15 (mostly 5-10 on average), if you are still experiencing too many artefacts at anything higher than those settings it is likely more light is required whether that be IR or LED flood. Default factory sharpness of 50 and 3DNR of 50 is way too high in my opinion, even sharpness of 25 is too much for my use. This is based purely on personal experience and also at shutter speeds from 1/250 to 1/1000 in an attempt as always to freeze motion

 

[Dragon]

[QUOTE="Thanks, I should be able to get something out of Computar hopefully, the IPC-HF8232F has an aperture of f1.2 so if that can be matched with a similar f1.2 lens which I expect to be around $100-$150 dollars equivalent it should do a better job in low light than the f1.4 and f1.53 apertures on the bullet cameras, all in theory/on paper of course ![/QUOTE]

I'm not super familiar with iris rules but the largest iris available for the Bosch 8000s is f1.5 to full close and the cheaper lens that looks like their more "default" lens is f1.8 to f8. Given that Bosch doesn't offer a lens lower than f1.5, I wonder if going lower than f1.5 has any significant benefit.

 

[CamCrazy]

I'm not super familiar with iris rules but the largest iris available for the Bosch 8000s is f1.5 to full close and the cheaper lens that looks like their more "default" lens is f1.8 to f8. Given that Bosch doesn't offer a lens lower than f1.5, I wonder if going lower than f1.5 has any significant benefit.

I believe this is purely linked with the resolution/cost, i.e. 5MP might have max f1.5 aperture and 2MP might have f1.2, one thing I can tell you for sure with all other things being equal is that I would take a f1.2 lens over an f1.5 all day long if that cameras sensor can match it. It seems the higher resolution sensors largest aperture is smaller than lower resolution cameras largest aperture if that makes sense! in other words the 5MP camera will let in less light than 2MP but this is pretty standard, hence if sensors are of equal quality the 2MP version should be better at night than the 4MP or 5MP and so on. No doubt a CCTV camera with 5MP resolution and f1.0 capabilities could be made but I wouldn't want to be picking up the cheque for it lens would likely cost as much as the camera

 

[Dragon]

The largest aperture lens Bosch has for the 2MP starlight is f1.4, though it's sized for a 1/2" sensor on a camera with a 1/3" sensor, so I assume that would be like putting a larger aperture than f1.4 would imply.

With 0 NR/sharp I feel like I'm seeing the image the camera is actually taking. It's significantly more blurry/hazy when you digitally zoom. Tweaking the lens focus doesn't help. License plates are readable, but blurry at 1/2000 (sky is overcast here, image is digitally zoomed):



1/4000 shutter doesn't help:


Here's the same lighting conditions, 1/2000 exp, 50 3DNR, 50 sharp:


That looks much better, but it also looks overprocessed. A couple frames later is an example of the "blurry block" I mentioned in a previous post:


I suspect the blurred areas happen because of this effect (0 NR, 0 sharp):


In the unprocessed image, it really looks like stream compression artifacts have clobbered parts of the license despite max quality and max bitrate settings.

Then here's 50NR, 0 sharp:


WTF? I don't get how the camera can get anything useful out of this blur by sharpening it. Photoshop sharpen tools sure can't. There must be more going on than 3DNR followed by sharpening to get the clear plates... Maybe it sharpens followed by 3DNR? Don't see that working either.

 

[CamCrazy]

The largest aperture lens Bosch has for the 2MP starlight is f1.4, though it's sized for a 1/2" sensor on a camera with a 1/3" sensor, so I assume that would be like putting a larger aperture than f1.4 would imply.

With 0 NR/sharp I feel like I'm seeing the image the camera is actually taking. It's significantly more blurry/hazy when you digitally zoom. Tweaking the lens focus doesn't help. License plates are readable, but blurry at 1/2000 (sky is overcast here, image is digitally zoomed):


View attachment 21768

In the unprocessed image, it really looks like stream compression artifacts have clobbered parts of the license despite max quality and max bitrate settings.

Dragon, if that bottom unprocessed image is the full size/resolution then something somewhere isn't right (stating the obvious I know!), lots of artifacts like you say and some odd processing - if you asked me without any prior knowledge of the situation I would say that is a sub stream at about D1 resolution or lower, that is how it looks at least. Is that camera at full zoom? what is the distance from camera to vehicles? Also is this issue only affecting this camera? and does the live stream look OK? twenty questions sorry! I think for testing purposes leave both 3DNR and sharpness at zero, in other words off. Thanks

 

[Dragon]

Dragon, if that bottom image is the full size/resolution then something somewhere isn't right, lots of artifacts like you say and some odd processing. Is that camera at full zoom? what is the distance from camera to vehicles? Thanks

Optical zoom is about 66% of max zoom. Digital zoom is way up (not sure how to quantify it). Distance from camera to license is 41.0833' measured with laser distance meter.

 

[CamCrazy]

Optical zoom is about 66% of max zoom. Digital zoom is way up (not sure how to quantify it). Distance from camera to license is 41.0833' measured with laser distance meter.

So you are using digital zoom? if so I would turn it off as soon as possible, it will do nothing but bad things for your image, optical zoom only, that will be 99% of the problem I would bet. If you are requiring more zoom then do not use digital, either move the camera closer or get a larger optical zoom something like the IPC-HFW8231E-Z5 should do a pretty good job on license plates, I intend to test one myself for that purpose at some stage. Just to clarify, turn noise reduction to 5 or 10, sharpness to 0 and digital zoom off or 0 as well