BoobieCam Installation for Face Detection in Front Entrance

[CanCuba]

My front entrance provides a perfect pinch point to get video of everyone who enters and exists the property. The entrance from the street is via an iron gate (currently uncontrolled but I will be installing an electric lock to be controlled by a VTO2211G-WP in the near future). The distance between said gate and front door is about 10m with 5 stairs at about the halfway point. This is really ideal to install a camera low, ideally camoflagued or hidden, to be able to face capture everyone who comes and goes.

So I decided to use the BoobieCam (Dahua IPC-HDBW5441F-AS-E2) which has two independent sensors in one cam with a normal footprint. The lenses can each be adjusted to almost any position the user requires. @looney2ns did an excellent review here.

The cam comes with either 2.8mm or 3.6mm focal length lenses. Since I am only really interested in small regions on each channel, I used the JVSG Lens Calculator to calculate the focal length for my needs. I ran through the calculations here and came out with 8mm for the gate side (being a bit farther than than the front door) and 6mm for the front door side.

I bought a refurbed BoobieCam off of @EMPIRETECANDY for the very low price of $95. Camera arrived in great condition and shipping was fast. First time being a refurb cam off of Andy and I can't be happier with his service and the condition of the camera.

The lenses were ordered off of Amazon and I went to work swapping them out. The first channel was easy. The sphere which holds the lens and sensor came apart easily and I was able to unscrew the factory lens and install the 6mm lens. Couldn't have been easier.

The sphere for the second channel was a bit more challenging. None of the three screws that hold the sphere together would budge. Having tried all my tricks (a bit of oil, duct tape to try and get a better grip, etc) I finally ended up drilling out the heads of the screws. This was a bit harrowing as I was worried about damaging the camera but all went well and I was able to swap out the lens on the second channel. To hold the sphere together without the screws, I used a two-part adhesive putty. This putty dries quickly and takes only a few hours to cure completely. I use it to install shower gel dispensers to shower tile where it constantly gets wet. It's great stuff and easy to work with.

When time came to mount it, I took down a camera which was mounted far too high (5m above the floor). Installed a keystone (female RJ45) connector and connected that to the existing cable and ran the new cable through flexible conduit and down to where the camera would be mounted. Installed a waterproof junction box and mounted the camera below that. Everything is installed above the footing of a column at a height of just over 1m. For getting faces of people looking downward coming out of the house, this height is really ideal.

I set the cam to the following on the camera GUI:

h.264H
8192 kbps
5 fps (this cam is underpowered for two channels and I had to drop from 15fps to 5fps to get both channels to detect correctly)
5 i-frame

Camera without cover. You can see the putty used to reassemble the sphere of the second channel.




Waterproof junction box and camera installed without cover.



And with the cover.



After focusing the lenses with the cover off and orienting them to get the desire coverage, we were off to the races! Or so I thought. Unfortunately, the longer focal length lenses seem to bring out more distortion caused by the glass domes. Others have mentioned slight image distortion issues with the factory lenses but the distortion was much more pronounced.

Door channel



Gate channel



Initially, I thought the lenses had been brought out of focus when the cover was installed. But after cleaning the glass domes and looking at the images, I could see that parts of the images were in focus and others slightly out of focus. Unfortunately, there doesn't seem to be an easy solution to this with the non-factory lenses. So we left things for a few days to see how images would come out.

So far, so good. Yesterday was the first full weekday with the camera installed and the channel facing the street produced 2135 face captures. Some of these are of the same person but I when I fine tune the face detection settings I will be able to reduce this, if I desire. Currently, the sensitivity to face captures is set as low as it will go and the time between detection is zero, or almost zero, which produces the duplicates. But storage is cheap and having more than one shot to identify someone can't be a bad thing.

To give you an idea, here's a capture of someone clear across the street, a distance of about 18m.





And someone walking right by on the sidewalk.






So the target distance which is really 5.5m (distance to the gate) is easily covered. The people on the sidewalk just outside the gate are about 6m away from the cam and I feel this is absolutely an identify quality image. The individual across the street is also easily recognizable, perhaps even identifiable, which I'm more than happy with.

The cop who was by today for another issue was very impressed with the results and said he's looking forward to having this resource available to him. It caught someone running with the knife early Saturday morning. The individual was in mid-stride at just after 2am. The photo is definitely identify quality but I can't share it here.

In closing, despite some distortion issues, which were brought on by me modifying the camera, I'm very happy with this project. The cam is picking up people on bicycles, motorcycles and the odd car passenger. The cam often picks up a face the first frame the person is in the image and it starts capturing. As mentioned, I had to drop the fps to 5 to be able to achieve this but it's hardly a trade-off considering I'm only interested in face detection. I have cameras covering the entire street and this cam was an addition to the system to be able to pick up faces of people coming and going. Getting approximately 90% of the people who pass by is just an added bonus.

One must be aware of the shortcomings of this camera:

1) Cam is under-powered for two channels. Setting the fps to the standard 15 may mean that some events (IVS, face detection, etc) may be missed, particularly on the second channel.

2) Turning the lenses too far out may result in some distortion. This could be overcome by setting the camera further back, if possible, which would mean the lenses are turned a bit further in.

3) Replacing the factory lenses may result in increased image distortion. This isn't so much a shortcoming of the camera but one should be aware of it.

4) The camera GUI requires a plug-in. While we won't get into the politics of this, be aware that one must be running Internet Explorer and be prepared to install a plug-in. I was able to setup the camera by using Firefox (without the plugin) while sitting in front of the TV which is hooked up to the NVR. Dialing in the settings was just as easy as ever. But be aware that I didn't need to draw any IVS lines which would be impossible to do with the camera GUI if one isn't able to install the plugin.

Pros:

1) Excellent face detection capabilities. I estimate we're getting 90% of the people who walk by right to left and 75% of people who walk by left to right on the second channel. For sure, this will be handy when an incident occurs.

2) Small footprint. Having essentially two cams in the footprint of one is ideal.

3) Excellent mic. Didn't buy this for the mic but it is almost as good as the mic on my Color4K/X. Those with a Color4K/X will appreciate what I'm saying.

I would recommend this cam (Andy has a few available as refurbs right now) to someone wanting to do the same. Since installing Friday, the cam has not dropped off nor required any intervention. Still, I may try a factory lens to confirm whether or not the longer focal length lenses are what's causing the distortion. But considering the quality of face captures the camera is producing, I may just leave it.

Hope this helps someone who would considering using this camera for a similar project. A couple issues to be aware of but I'm very happy considering I paid $95 for what's basically a brand-new cam that retails for double what I paid.

 

[actran]

@CanCuba I'm curious how well the camera built-in face detection works for you at night, especially with the dome cover on.

How often does this camera miss a face detection? (Day vs night)

The camera built-in IR seems to cause image issues on my side so I'm forced to turn it down low or off.

 

[CanCuba]

I am getting decent face detection at night.

For people walking right to left, I'd say I get > 90% during the day. At night, it may be slightly lower. Say 75% to 80% but the image quality isn't as good, which is to be expected.

I have the camera set to colour at night and have my powerful 200w LED panels illuminating the street.

I'll pull some night samples tomorrow to demonstrate what I'm getting. It's still very good for people on the sidewalk but the quality deteriorated quite quickly for targets farther away.

But it's still surprisingly fast to catch faces. The capture of the knife wielding target from a few nights ago was very decent and this guy was almost in a full sprint when he passed the camera. Can't post that one (yet) unfortunately.

 

[CanCuba]

@actran Here you go, my guy. Not great shots but I'll take them.

Just to give you an idea of how fast faces are detected, here's a guy on a motorcycle travelling about 20km/h, at night no less. Camera snaps a pic the second frame he appears. Not sure why the video starts at the point of detection.

I'll see if I can find one of someone walking where they're detected first frame.





View attachment NVR_ch15_main_20230704221125_20230704221135.mp4


















From about 12m:









So these night captures are just "okay". I don't really have an IVS use for this cam so I just leave it on face detection and whatever it gets is fine with me. Could probably do some fine tuning which would mean washing out the column which isn't a big deal. I'm just worried about people coming through the gate at night and getting a clear face shot. That is my goal, after all.

Not having any experience in cleaning up photos, I'm curious how much someone could clean up the results. I'm NOT adding more light as I already have plenty in this location. lol

 

[CanCuba]

@actran

Forgot to post someone on the sidewalk. This is more the range that I'm interested in.

 

[actran]

Thanks @CanCuba

#1: Do you mind sharing links to the lenses you bought?

#2: Are you using any software to manage all the snapshots? For both viewing and tagging.

#3: On camera "Face Detection" configuration page, what are your thoughts on each option like "face enhancement"

 

[CanCuba]

Thanks @CanCuba

#1: Do you mind sharing links to the lenses you bought?

#2: Are you using any software to manage all the snapshots? For both viewing and tagging.

#3: On camera "Face Detection" configuration page, what are your thoughts on each option like "face enhancement"

1)

https://www.amazon.ca/gp/product/B01MDTOU9C/ref=ppx_yo_dt_b_search_asin_title?ie=UTF8&psc=1

2) I use an NVR right now. It manages all the snapshots but isn't saving to the proper resolution for some reason. But I have an SD card in the camera and I log into the cam and download the images in the proper resolution.

3) I use Face Enhancement and Target Face Brightness. Target Face Brightness causes the video to "strobe" as it enhances the face capture but that isn't an issue for me. I also have the sensitivity as low as it will go, the face angle up to 90 degrees and the detection time as low as it will go. Basically, I'm trying to capture all the faces I can. The last two days, the first weekdays with the camera mounted, I'm getting about 2300 snapshots per day.

The camera has captured up to 4 faces at the same time (I'll try to get an example later) and has captured both people on a moving motorcycle. As I've mentioned, I'm quite impressed with the face capture abilities.

I do have the fps set at 5 which I think gives the cam more time to analyze each frame and pull out the faces.

The Color4K/X runs face detection when it's on daytime settings (currently 7am to 8pm). It gets about 500 faces a day but it's a bit farther back and high (7m above the street level). It has a 2.8mm lens and is used as an overview but I'm happy with the daytime captures.

Also, the SD1A404XB runs face detection 24/7 and gets a few more faces than the Colour4K/X as it's a bit lower and closer to the street. Nighttime captures are decent. Other cams are doing the IVS so now having 3 cams doing face detection on the street during the day gets me virtually everyone who passes.

Going to install 2 pinhole cams on the outside of the column visible in the BoobieCam gate view. They'll be about 4' off the ground. Having seen how many faces I get with the BoobieCam, the pinhole cams should get me > 99% of everyone walking by.

Personally, I think face detection is very useful in my scenario which is mostly foot traffic. There's almost no theft done with vehicles involved (as of yet) so getting pedestrian targets is ideal in this situation.

 

[dmcIPCAM]

Where else can you see a Polski Fiat 126 in the wild! - This is why I love screenshots from @CanCuba

When you are packing a 704cc twin under the boot, you definitely need the Lotus green paint job and the racing stripe.

Joking aside, it's amazing how mechanically resourceful folks are on that island!

 

[CanCuba]

Good eye! lol

There's actually two of them, both green, who park there daily. The owners work in a government office on the other corner. Other one has "chrome" air intakes where the ones on the car in the pic are black plastic.

I've only been in one and it had a larger Toyota motor. They really are clown cars when you see 4 adults climb out of one but they get the job done!

 

[CanCuba]

@actran

Here are some captures from moving vehicles.

Same couple on a scooter:










View attachment NVR_ch15_main_20230705145433_20230705145444.mp4


















Passenger in a van:





View attachment NVR_ch15_main_20230705162416_20230705162425.mp4


















Kid in a delivery trike (face blurred out as it's a kid):





View attachment NVR_ch15_main_20230705175556_20230705175607.mp4

 

[CanCuba]

@actran

And my settings for face detection. As you can see, I have quality threshold, etc turned all the way down to maximize the number captures.

Also, not all cameras, such as the SD1A404XB, do not support Face Exposure. It seems to be a more advanced feature.

 

[actran]

@CanCuba Thanks, impressive given the modified camera you had to work with.

 

[CanCuba]

@CanCuba Thanks, impressive given the modified camera you had to work with.

Thank you.

I'll post a picture later of how we tried to camoflague it. Didn't turn out the way I'd like but it's still less obvious than just having the camera and junction box in plain sight.

We considered having an African mask designed and 3D printed to hide the camera. Each dome would have been an "eye" of the mask. But having something custom designed is a bit pricey and this cam does run a bit hot. Hasn't been an issue yet with the blast of heat we've been getting so I think it's getting enough air flow.