Detecting wildlife with Dahua cameras

onthewater

n3wb
Joined
Oct 29, 2018
Messages
5
Reaction score
6
Location
CA
There might be a reason why all the 'game cameras' sold at sporting goods stores for hunters usually just use PIR. The issue is the 'motion' goes off for anything, not just 'animals'.

I work in the computer field and there are a lot of upcoming AI things happening, including things like the Hailo and Coral TPU. These hardware accelerated machine learning devices could likely take your video input and process it and do the actual animal detection on their chips, rather than relying on the camera itself.

While animals are a fun thing to 'track', I think the security manufactures are making a lot more money on human and vehicle detection.

Just some thoughts, just my .02 cents, just some ideas...
 

tigerwillow1

Known around here
Joined
Jul 18, 2016
Messages
2,997
Reaction score
6,150
Location
USA, Oregon
While animals are a fun thing to 'track', I think the security manufactures are making a lot more money on human and vehicle detection.
You're absolutely correct about where the money is made. For me and others, detecting the animals is more than fun, because in some settings animals are the main threat and cause of damage. We never would have known what happened to one of our cats had it not been for the video of a coyote nailing it. I've had a few cases of identifying how barriers around crops were breached, leading to hardening the defenses. I totally support Dahua enhancing the cameras to better serve the needs of their customer base. My beef is that the blasted human/vehicle detection can't be turned off, and plain-old nice but stupid IVS has been eliminated from the cameras with the better image sensors.
 

jrbeddow

Getting comfortable
Joined
Oct 26, 2021
Messages
270
Reaction score
358
Location
USA
You're absolutely correct about where the money is made. For me and others, detecting the animals is more than fun, because in some settings animals are the main threat and cause of damage. We never would have known what happened to one of our cats had it not been for the video of a coyote nailing it. I've had a few cases of identifying how barriers around crops were breached, leading to hardening the defenses. I totally support Dahua enhancing the cameras to better serve the needs of their customer base. My beef is that the blasted human/vehicle detection can't be turned off, and plain-old nice but stupid IVS has been eliminated from the cameras with the better image sensors.
Wow, thank you for this comprehensive thread that fully explains the limitations imposed on the newer firmware of many Dahua cameras, especially the 5442 series. I had suspected that the IVS doesn't really shut off the vehicle/human detection (when unchecked) on any of the recent firmwares that I have tried, but you have confirmed that, as well as confirming that it used to work on older revisions. Nice work, and I agree that it is maddening. Is it worth trying to point this out to @EMPIRETECANDY and @Wildcat_1 to see if anything can be done to re-enable these controls?
 

tigerwillow1

Known around here
Joined
Jul 18, 2016
Messages
2,997
Reaction score
6,150
Location
USA, Oregon
Is it worth trying to point this out to @EMPIRETECANDY and @Wildcat_1 to see if anything can be done to re-enable these controls?
Already done, and it didn't need any pointing out to Andy. He has been very helpful, gracious, and transparent with me about this. My own conclusion is that nothing will change unless and until a big customer cares about it. I'm still doing some testing and plan to report results in a week or two. It goes slow, the main reason is I haven't been able to train wild animals to walk through the tripwires on my command :)
 

jrbeddow

Getting comfortable
Joined
Oct 26, 2021
Messages
270
Reaction score
358
Location
USA
Yes, currently maybe use the none AI model IPC-T2231T-ZS still can catch the animals, other models has less false alarms on animals now.
With all due respect, animal crossings should not be considered "false alarms" in this context: all it would take is a way to truly disable the filtering that takes place for people and vehicles when those checkboxes are unchecked. Unfortunately, those controls don't do anything anymore.
 

Wildcat_1

Known around here
Joined
Dec 9, 2018
Messages
1,905
Reaction score
5,258
Location
US
With all due respect, animal crossings should not be considered "false alarms" in this context: all it would take is a way to truly disable the filtering that takes place for people and vehicles when those checkboxes are unchecked. Unfortunately, those controls don't do anything anymore.
@jrbeddow I do have an idea on this one. SMD 4.0 for example features (on the latest cams of which only a few have this currently) increasing detection around a wider set of AI animals. This is used to determine what NOT to trigger on of course. My thought here is that we could use that on an inverse. In other words we paint the picture of where this may be needed (use case) and that based on that look to use the AI algorithm the other way. This is to say that when human/vehicle detection is unchecked look for targets that are NOT human/vehicle to trigger upon and use the AI algorithms that detect animals to NOT cause a trigger, to , in that example, trigger a captured. Might be a way to approach that as a need as new cams are released. Will think on that some more but just something that came to mind as I was playing with SMD 4.0
 

tigerwillow1

Known around here
Joined
Jul 18, 2016
Messages
2,997
Reaction score
6,150
Location
USA, Oregon
I want to make sure I'm not accidentally leaving the impression that I want the cameras to detect ONLY animals. I want to know about the people and vehicles going by my house as well as the coyotes, cougars, etc. Out of curiosity I compared firmware file sizes. The 4431-series cameras do everything I want except for their image sensor not doing well in the dark, firmware file size 14 megabytes. The 5442 and 5842 camera file sizes are now around 86 megabytes. and they lost the ability to detect animals. I know, I know, the majority of users don't see this as a loss, it just feels needlessly dumb to me.
 

Wildcat_1

Known around here
Joined
Dec 9, 2018
Messages
1,905
Reaction score
5,258
Location
US
I want to make sure I'm not accidentally leaving the impression that I want the cameras to detect ONLY animals. I want to know about the people and vehicles going by my house as well as the coyotes, cougars, etc. Out of curiosity I compared firmware file sizes. The 4431-series cameras do everything I want except for their image sensor not doing well in the dark, firmware file size 14 megabytes. The 5442 and 5842 camera file sizes are now around 86 megabytes. and they lost the ability to detect animals. I know, I know, the majority of users don't see this as a loss, it just feels needlessly dumb to me.
Yes understood which is why I was theorizing that utilizing AI to assist in animal detection only when Human/Vehicle is disabled could be a way to tackle it IF the use case could be justified.
 

Mike A.

Known around here
Joined
May 6, 2017
Messages
2,669
Reaction score
3,884
After also messing around a lot with it, seems like what's happening is that the AI model still is active with IVS on no matter how you have those controls set. I suspect that something along the lines of what you're saying is happening - the model continues to operate with some discrimination against animals vs more simply as it did before.
 

Wildcat_1

Known around here
Joined
Dec 9, 2018
Messages
1,905
Reaction score
5,258
Location
US
After also messing around a lot with it, seems like what's happening is that the AI model still is active with IVS on no matter how you have those controls set. I suspect that something along the lines of what you're saying is happening - the model continues to operate with some discrimination against animals vs more simply as it did before.
Exactly and thats why I mentioned, 2 key pieces could be pursued, 1) ensure AI algorithm is disabled when Human/Vehicle/Non-Motor Vehicle (in case of 7x series) is unchecked, 2) use SMD 4.0 AI algorithm for inverse target capture focused on those needed animal capture. I.e.. when AI algorithm is disabled (therefore no human or vehicles are being targeted) instead flip to using the SMD 4.0 'dismiss animal's intelligence to instead target those dismissed objects. That way you still benefit from things like dismissing tree's, branches, leaves, blowing trash etc. Now I just have to build the case, document how to do it in code and sell it to Dahua for their feedback :)
 

tigerwillow1

Known around here
Joined
Jul 18, 2016
Messages
2,997
Reaction score
6,150
Location
USA, Oregon
That way you still benefit from things like dismissing tree's, branches, leaves, blowing trash etc.
If that could be achieved it would be the best of both worlds. I'm currently running a 5231 and 2231 side-by-side. The 5231 gets a lot more triggers. The 2231 often misses small animals like mice or chipmunks, but it also misses bugs flying through the IR light, shadows, webs, and so forth. I'm torn over which way I like better. My 4231 LPR camera even triggers on scorpions within a few feet of the camera.
 

quest100

Pulling my weight
Joined
Aug 10, 2018
Messages
90
Reaction score
109
Location
CA
If I remember correctly the Dahua cameras have a pair of wires for external triggering. Does anyone make an external PIR sensor that can easily be used to externally trigger a Dahua camera? Plug and play with adjustable sensitivity, battery powered, and no soldering required?
 

tigerwillow1

Known around here
Joined
Jul 18, 2016
Messages
2,997
Reaction score
6,150
Location
USA, Oregon
If I remember correctly the Dahua cameras have a pair of wires for external triggering.
Good out-of-box thinking! I'm kind of a debbie downer and thinking of the downsides first. With not too many exceptions, the turret cams don't have the alarm input, and many of the bullet cams don't. Since I run external IR illuminators, there's probably sufficient power available for a PIR sensor. I'm thinking the biggest obstacle would be the range of the PIR sensor, which I'm guessing is limited to ~30-40 feet. IVS picks up things a few hundred feet away. If I had a 32 channel NVR instead of 16 channel, I'd be experimenting with side-by-side cameras, a dumb one to detect something is there, and another with a better image sensor see what it is. A lot of things that are pretty well visible on a 5442 or 4k-x are just a moving blur on the other cameras.
 

Mike A.

Known around here
Joined
May 6, 2017
Messages
2,669
Reaction score
3,884
That's easy to do in BI. Just set up a group that gets triggered from one or more other cams. That's what I do in my backyard. I have a 5231, 2231, 5442 and the 4K-X. If any of the cams trigger, they all trigger. So it does as you describe. The IVS on the 5231/2231 say something's there and trigger the 5442/4K-X. I also have a motion detector that does the same (which is what usually triggers things before anything else).
 

tigerwillow1

Known around here
Joined
Jul 18, 2016
Messages
2,997
Reaction score
6,150
Location
USA, Oregon
The IVS on the 5231/2231 say something's there and trigger the 5442/4K-X.
Wow, you're way ahead of me. All I'd need is a new NVR (Dahua NVR can also trigger multiple channels on an event from one camera), new switch, a few k$ more of cameras, and a place to hide from my wife :) I'm already on the edge of the cliff talking about the new PTZ Andy posted info for a few days ago.
 

tigerwillow1

Known around here
Joined
Jul 18, 2016
Messages
2,997
Reaction score
6,150
Location
USA, Oregon
Side-by-side comparison of 5231 varifocal, 2231 varifocal, and 5442 6mm fixed lens cameras over a month.

Disclaimers:
Almost totally centered on detecting animals and their image quality. Many other camera features ignored.
Conclusions and options are observational, not based on a controlled study.

The cameras:
IPC-HDW5231R-ZE varifocal firmware V2.800.0000016.0.R 5-16-2020 (pretty old)
IPC-T5442TM-AS 6mm fixed lens firmware V2.800.00000008.R 9-2-2019 (VERY old)
IPC-T2231T-ZS varifocal firmware V2.820.15OG001.0.9 9-1-2021 (fairly new)

An external IR emitter was used, and the cameras' IR lights shut off. Shutter speed for night profile is 1/50. All sample images are digital zooms, with an attempt to make the animals about the same size for comparison.

The 5231 and 2231 were zoomed to the 5231's narrowest FOV of 34 degrees (2231 can go narrower to 30 degrees). The
5542's FOV is 56 degrees. The setup and setting, the wider image is obviously from the 5442, and it's 300 feet to the far end of the driveway:
CameraSetup (Large).jpg FullView-2231.jpg FullView-5442.jpg


Comparing the 2231 against the 5231:
The only reason I tried the 2231 was its much better light sensitivity spec of 0.002 lux vs. the 5231's spec of 0.006 lux. When first looking at the 2231's image it appears to be far superior to the 5231's, with better contrast, more vibrant colors, and significantly less video noise at night. When comparing digital zoomed animal images however, I found no difference in being able to identify an animal, so for the purpose of animal identification, I call the video quality a wash, and I'm really suspicious of the comparative lux specs.

There was a real difference in animal detection capability using tripwire. Both reliably detect animals the size of a typical cat or larger. The 5231 reliably detects very small animals (mouse, chipmunk, small birds). With the 2231, it's hit-and-miss, 50% probability I'd guess. With this reduced detection sensitivity comes reduced false positives. The 2231 pretty much never trips on insects flying through the IR beam or webs in front of the lens, which the 5231 often triggers on. Which is the better choice? If you want audio, it's the 5231, end of discussion. For a more pleasing to look at image, the 2231 wins. Detection preference could favor either, depending on if you favor very small animal detection or fewer nuisance trips. I'm not posting any test images because I don't think there's any significant overall difference. (There is another 2231 model, a bullet with an external audio input, the IPC-HFW2431T-ZAS-S2 which I was unable to find a non-vaporware source for).

Comparing the 5442 against the 2231 (and indirectly, the 5231):
Image-wise, during the day, digital zooms from the 5442 are better in every situation with a small exception approaching 300 feet from the camera, where the optical zoom of the 5231 gives a small edge. At night, the 5442 is clearly better as the useful IR light does not reach even half way to the end of the driveway. The 5442 and 2231 have the same lux sensitivity of 0.002, leading me to suspect the 2231's spec is not correct.

The 5442 falls down on very small animal detection. It's still fairly reliable with the typical cat size or larger, but unlike my description of the 2231's very small animal detection of about 50%, with the 5442 I'd peg at as about 10%, or almost never. I started this exercise for replacing a few of my older, lower light sensitivity cameras. Of these three, I'll pass on additional 2231s, probably add a couple more 5442s, and keep the 2 megapixel starlights. I was hoping to replace all my fixed-focal chinese market 4331s, but will keep a couple of them as they are very good with animal detection, except for the ones they don't see at night with their lower-end image sensor. Then I'm out of the new camera market unless and until Dahua allows human/vehicle detection to be shut off. The 4k-x would be a killer camera for me if it would detect the animals. Remember, if you get a new 5442, it's the fixed focal version only, and I don't know if the -ASE version will take the old firmware. I can say from this exercise that the human/vehicle only detection in the newer firmware unfortunately works very well.

The sample images are 2231 on the left, 5442 on the right.
The first 2 sets show objects 300' from the camera and show that using greater digital zoom with the 5542 competes well with the 2231's maximum optical zoom:

Car-2231.jpg Car-5442.jpg
wheel-2231.jpg wheel-5442.jpg

With the next 2 sets, the squirrel triggered the 5442's IVS, while the quail family didn't:

squirrel-2231.jpg squirrel-5442.jpg
Quail-2231.jpg Quail-5442.jpg

The mouse and rabbit were not detected by the 5442's IVS:

Mouse-2231.jpg Mouse-5442.jpg
Rabbit-2231.jpg Rabbit-5442(not detected).jpg

This rabbit did trigger the 5442's IVS:
Rabbit1-2231.jpg Rabbit1-5442(detected).jpg

Finally, this deer is about 100' from the camera, illuminated a bit by a security light, and triggered the 5442 IVS:
DeerWithLight-2231.jpg DeerWithLight-5442.jpg
 
Top