It sounds like it is working as intended.
You will find that AITOOL at the moment has way more granular features and options than the integrated Deepstack with BI.
The BI integration is more about the simplicity of operation over the fine detail granular level some of the 3rd party
tools provide. But the 3rd party is usually harder for a novice to set up as well. The BI integration allows a novice to set it up with a download and few check box clicks and not having to deal with Dockers, etc.
Do not get me wrong, it is no slouch and has come a long way since it was first introduced, but many people coming from the 3rd party end up going back because of the more detailed settings you can set in something like AITOOL.
The time for motion trigger in BI has absolutely nothing (well mostly nothing) to do with how long deepstack is engaged during that event, that is set elsewhere. It is totally possible to have the trigger occur longer than Deepstack will analyze.
In the AI setting for the camera is the option to tell it how many images to send and at what time frequency (pre-set pull down options). It is totally possible that Deepstack analyzes the first 4 images and determines nothing is there and doesn't send the rest of the images and then comes up with the Nothing Found. This can happen with headlight shine that causes a camera trigger but the car is still too far away, so deepstack analyzes the first few frames and if appears all is stationary, it cancels.
You can get around that by putting in the to cancel box something like Zebra or Banana or Toilet. That will force Deepstack to analyze EVERY image for the number of images and at the time interval you select and then if it finds something, it will use the best picture for the alert image. But this can drive CPU up because if it finds the object in the 4th image and you tell it to analyze 20, it will still go thru all 20. Depending on your system, this could also cause Deepstack to timeout and then show nothing.
Here is an example of how the To Cancel option works. At night, my one camera, that has a straight on angle of the street to get a side profile of a car, 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.
Once I added a banana in the cancel field, it now will go thru all the images and select the best one, which gives me the whole vehicle in the frame. It makes for scrubbing video much quicker as I can skip looking at video of known vehicles.
Many people have ran both simultaneously and found that they performed the same and caught the same triggers, but the BI integration was easier to set up so they went with that. Others that ran both at the same time said the 3rd party was more accurate so they went back to it. It is field of view and system specific. YMMV.
It sounds like you have a problematic field of view for the BI integration and if the tips I provided still don't get you the results you need, then you probably need to stay with the 3rd party tool until the BI integration matures more. But for most of us, this is spot on and uses less resources and a lot simpler than a 3rd party.
I hope this helps!