False alarms

[graemewolfenden]

My Accusense camera alarms are constantly being activated in the middle of the night by next door's cat. He is quite a bit boy but nowhere near the size of a human which is what they have been set to detect. I read about complicated algorithms etc but mine don't seem to work.

 

[Dan]

Hi @graemewolfenden

Can you confirm the camera model you have and the firmware version it is using?

Are you recording to an NVR? If so what model of NVR and what firmware?

Can you send us a screenshot of the settings for the smart event you currently have set?

 

[JB1970]

Yup. I get cats (and hedgehogs) detected as human targets with the Acusense 2386G2 and the 2346G2.

 

[graemewolfenden]

My cameras are DS-2CD2346G1-I/SL running V5.6.0 build 190507

My NVR is a DYN 7608NI running V4.21.005 build 190509.

I've attached a screenshot. In case you are trying to ascertain if I have ticked "human" - yes I have

cheers

 

[JB1970]

It looks like you’re trying to cover a lot of individual areas there - Id maybe have started with a single line. A couple of potential problems there:

1- Minimum and maximum object sizes aren’t set

2-Line crossing works best, set to detect objects moving left/right across a vertical line rather than toward/away from the camera

2-The lower line (yellow) is too close to the edge of the field of view. The object needs to be in view before it crosses the line. The camera needs to determine whether or not the object is between the minimum/maximum size, moving in the correct direction, perceived to be human, and that the appropriate proportion of the target has crossed the line/entered the intrusion area. It can’t do that if the target is already in the line as it appears in view.

3-The sensitivity is an inverse of the target size. IE: if the sensitivity was set to 25, then 75 percent of a valid target needs to cross the line/enter the intrusion area. So 100 is not a good value to start with as 0 percent of the valid target needs to cross the line. Start at 59 and adjust from there.

One more thing to be aware of. The AcuSense function works for pre defined “Smart Events” as you have marked in your screenshot. If you do a search for activations with “Smart Playback” and select a different area, it will return results of all targets, human or not as those results come from the NVR searching the metadata embedded in the recorded video (if dual VCA is enabled in camera)

 

[graemewolfenden]

It looks like you’re trying to cover a lot of individual areas there - Id maybe have started with a single line. A couple of potential problems there:

1- Minimum and maximum object sizes aren’t set

2-Line crossing works best, set to detect objects moving left/right across a vertical line rather than toward/away from the camera

2-The lower line (yellow) is too close to the edge of the field of view. The object needs to be in view before it crosses the line. The camera needs to determine whether or not the object is between the minimum/maximum size, moving in the correct direction, perceived to be human, and that the appropriate proportion of the target has crossed the line/entered the intrusion area. It can’t do that if the target is already in the line as it appears in view.

3-The sensitivity is an inverse of the target size. IE: if the sensitivity was set to 25, then 75 percent of a valid target needs to cross the line/enter the intrusion area. So 100 is not a good value to start with as 0 percent of the valid target needs to cross the line. Start at 59 and adjust from there.

One more thing to be aware of. The AcuSense function works for pre defined “Smart Events” as you have marked in your screenshot. If you do a search for activations with “Smart Playback” and select a different area, it will return results of all targets, human or not as those results come from the NVR searching the metadata embedded in the recorded video (if dual VCA is enabled in camera)

Thank you - quite a lot to go on there! Regarding there being a lot of individual areas - I do have three other identical cameras that have far less trigger events, but they also detect the cat!!!

A few points/observations on your comments.

1. Is it necessary to set minimum/maximum sizes when the "human" box is ticked?

2. Point taken - I will alter this.

2? Ditto

3. I will amend this also

 

[JB1970]

Thank you - quite a lot to go on there! Regarding there being a lot of individual areas - I do have three other identical cameras that have far less trigger events, but they also detect the cat!!!

A few points/observations on your comments.

1. Is it necessary to set minimum/maximum sizes when the "human" box is ticked?

2. Point taken - I will alter this.

2? Ditto

3. I will amend this also

I'm not sure how good the human/vehicle algorithm is - my wild guess is it's just looking for a basic pixel group moving at a particular speed, predictable direction to ascertain what it is. For instance a lot of times I get cars passing my house right to left in bright light and my car is on the drive parked at 45 degrees to the field of view. A line running across my drive was frequently triggered by passing cars, as they caused a reflection in the side windows of my car to move through the defined line top to bottom.

The max/min size is not essential but it may help with false detections. It can be awkward to get right as you can only set them globally for line crossing and a person traversing a line in the near field of view appears far larger than a person in the far field of view. I would set the min size based on a human at the line furthest away from the camera and the max size at the line nearest.

The newer G2 AcuSense cameras perform better. Strangely I've found that motion detection with the human filter gives far fewer false alerts than line crossing with the human filter.