For most of my cameras I have different day and night time motion settings in BI.
Daytime with Edge Vector will knock out a lot of false stuff. At night, because it is darker, edge vector can this miss a lot of stuff, so I use simple for most of my cameras at night.
Min object size is so camera field of view dependent. I have done this enough I can usually estimate based on what the field of view looks like. The best way is to have someone stand out in the camera view and then draw a box around them.
Do you want alerts for every car and person in the street or just on your property? That makes a difference in the min object size. This would be about the size if you wanted people in the street:
This camera is more of an overview camera, so you could probably make the make time 1.0 seconds and not miss anything and further knock out a lot of false from the tree.
I have tested my camera field of views with IVS versus Deepstack and in most instances, using the camera AI has resulted in better success. Of course YMMV and it really is dependent on the field of view.
For example, for a tightly zoomed in camera like a Z12E for plate reading duty (LPR), the camera AI sucks as the camera needs time to identify the object, determine if it is something you want, compare it to your IVS rule and then decide to trigger or not. These tight views and it will miss them, especially at night.
I run a few cameras with DS because for that particular field of view, it was providing better results.
Here is an example. At night the camera AI would struggle with this tight view. It has a straight on angle of the street to get a side profile of a car and it would miss it a lot of times because the vehicle just isn't in the field of view long enough, so this is a great candidate for DS.
Now the issue I had with DS is that it would either find a car but the alert image would be the lightshine on the street or just a part of the vehicle, or it would trigger out nothing found due to headlight bounce off the street.
DS has a "to cancel" option, which means it will analyze EVERY image to determine if the item is in it. Once I added a cancel banana in the field, it now will go thru all the images and select the best one, which gives me the whole vehicle in the frame and it eliminated the nothing found as well. It makes for scrubbing video much quicker as I can skip looking at video of known vehicles.
So even though I am a fan of camera AI, there are instances where DS will be better, or even BI motion detection. I use each one of them depending on the field of view and what is to be accomplished with that field of view.
Here was a thread started where I gave more details as to the AI in the camera:
It has come to my attention that certain Dahua cameras come with built in AI that distinguishes between human & cars. I was unaware of this. Well, always saw the AI word used in the reviews but never really thought about it because I thought it was a Dahua NVR integration only. I am currently...
ipcamtalk.com