another Dahua AI question

[Raylo32]

I am really liking the AI human detect performance on the rear deck Dahua so far... so I was wondering how it would perform in totally opposite type of scene that I have up front. Situation is a townhouse with the cam mounted in the upper corner of the garage opening facing a street... not a main street but several hundred cars a day going and coming out of the 'hood. And at night the traffic flow puts headlight wash onto my lawn and driveway almost up to the garage... it is partially blocked by the outside stairs up to the main level.

The question I have is: Can the IVS reliably ignore cars and their splashing headlights? I see it has a car detect setting, but I am needing just the opposite. In BI I mask the scene to block cars driving by and night time headlight shadow detects but to screen out falses I have almost no detection zone left.

This is the scene. The roads in the parking lot heading to and coming from the stop sign are 1-way. IOW the cars come in on the left, then make a left down the street in front of me which is 2-way. Has anyone deployed one of these cams in a scenario like this?

 

[wittaj]

Yes it will do just that. If you set it for human only, which it sounds like you want, then it will work just fine. Now I will tell you that with that truck there, it will probably struggle with people at night because the truck takes up so much of the field of view and if running IR, a lot of IR bounce will be coming off the truck, but it still may do it. Maybe bring the back camera around and test it tonight?

The true test....I have found the AI of the cameras to work even in a freakin blizzard....imagine how much the CPU would be maxing out sending all the snow pictures for analysis to Deepstack LOL. My non-AI cams in BI were triggering all night. This picture was ran through Deepstack (without the IVS or red lines on it) and it failed to recognize a person in the picture, but the camera AI did. This camera also gets the headlight shine you talk about and I never get any false triggers because of it. This pic says it all and the video had the red box over it even in complete white out on the screen:

 

[Raylo32]

Wow, Now that is cool. My goal with this front would be human only and even then I'd mask off the sidewalk, at least in day profiles, to avoid passersby causing alerts. One small problem area is the neighbor's driveway. The LOS to that gets a part of my stairs, but probably not enough to be a problem if masked... although masked just in daytime when the kids are playing.

 

[wittaj]

I think you could do an intrusion box along the steps to avoid your neighbor driveway and still get your triggers.

 

[bigredfish]

Yep, Human-only will ignore vehicles...... well, MOST vehicles... unless it sees a human hanging on it...

View attachment HOA Ent_ch1_20201026140306_20201026140330.mp4

 

[wittaj]

My 1985 analog PTZs with AI caught this surfin dude


Not really lol, but that is what I thought of with @bigredfish video LOL.

 

[Raylo32]

Or MJF on his back to the future hoverboard! I really want one of those. As for he garbage trucks, we get them twice a week and that surely would send human detects... and wouldn't be a problem using full field human detect... it would be a good routine system op check. My problem would be the couple dozen pedestrians that walk by every day causing nuisance alerts. I'd have to mask the sidewalk off or use other features like the intrusion box (that I haven't played with yet).

 

[Raylo32]

And here is an example of one version of the headlight issue... I recently redid the motion setup for this HIK cam in BI and it looks like I may have missed a bit of masking either behind the stair rails or on the lawn. I had 4 alerts last night with the lights in exactly the same spot. And this is just one mode of headlight nuisance. The other main one is when the truck isn't there and the light flashes over onto my driveway... from cars going both ways. It comes even beyond the threshold to the stairs, which is one of the main areas of interest. It is a real PITA. And headlights vary greatly, from the subdued ones in this photo to misaligned ones or some on high beam that spread light even more far and wide. I am considering moving the cam to the other side of the garage opening which might help some. Or I might get another Dahua and off load the motion detect from BI to the Dahua's native AI, like I am doing in the rear.

 

[wittaj]

Knock on wood, my Dahua AI check boxes are spot on in all of my cameras that have them! Has made scrubbing alerts in BI a breeze because even with how great motion detection is in BI, there are a few situations where I cannot knock out false triggers, especially at night with headlights bouncing off a hill for example. Trying to eliminate that and then I miss a real trigger. The camera AI doesn't even flinch at attempting to think the headlight bounce is a trigger, or motion lights turning on.

There are certainly things you can do with motion detection in BI to help knock out the false triggers from headlights with object detection checked and resetting based on a certain % of the screen getting motion (ie headlights), but you then could miss a real event. Also make sure edge vector is selected.

Other option is upgrade BI to the current with Deepstack Integration. Still a work in progress with updates all the time, but looking at your view, it is better than mine with a "dumb" overview camera I am using the integrated Deepstack integration for and it has been spot on.

 

[cctv-dave]

Your garden light is potentially affecting this more than the headlights, and being in colour, and not using HLC.
The cam angle is great for protecting the truck but not so good for doing the AI. Appreciate you're looking for a compromise but sometimes it takes more than one cam.

And that sometimes has advantages too..
I have 3 cams on the front of my property doing different jobs. 3 sets of IR means the image is improved on each cam than it would be with only 1 cam.
A happy benefit.