Storms - Lightening and Rain

[gnalsa]

Hi Everyone,

I have noticed that flashes of lightening are detected as 'people' objects and events are triggered/recorded.

Has anyone else seen this behavior? Any workaround?

Here are some strange items detected as an event:
- Strong rain
- Flashes of light (lightening)
- Flood light turns on/off

[Steve]

Hi Support

I'm also seeing issue in the new algorithms in version 4.1 as I have a rule people inside boundary that has been running for 3 years and all updates but now in version 4.01 the same rule, same camera when I turn on my car lights in the morning they shine on my drive and sighthound thinks its a person. so I'm getting false alerts, that I have never received before. this is causing an issue as I have a rule via IFTTT, that after 11pm if person triggers this rule beep an alarm. it has never given me false alerts and now it was waking me with car lights at 3am!!!!!. this is my main intruder alarm and making it impossible to use.

[Steve]

I have submitted a Bug Report

[gnalsa]

Bump - has anyone else seen this behavior? Any recommended work arounds?

[Semper Vaporo]

Various sudden lighting changes are detected as motion. Work-arounds are to change the detection rule somehow.

Or if you want detection of any motion in the whole view, but car headlights shining on a wall are detected as motion, define a convoluted window that encompasses the whole image, with an excursion from one edge to a box around the wall area where the lights show (sort of a box within a box or a keyhole shape extruded from one edge into the outer box)

I have to alter my motion detection rules a couple of times per year because summer shadows are different than winter ones due to the angle of the sun.

I have found that using "Crossing a line" is a good method of reducing false detections due to rapid lighting changes (lightning, car headlights, etc.). A couple of my scenes make the Crossing a line feature good to use... I put the line directly over a tree in the yard. Cars in the street pass behind the tree and most of the time they are not detected, as they appear to be an object that disappears behind the tree and a new object that starts on the other side. A few times the first object seemed to blend with what came out from behind the tree that appeared to pass in front of the tree, and thus Crossed the line.

Another thing I have done is tip the camera to get the most common motion in the scene to be horizontal across the scene. The scene is my front yard from a corner of the yard. Because of the location of the camera the street runs diagonal in the scene and car headlight lengthen the size of the object and this puts the "bottom" of the object lower in the scene than the real object ever goes. I had an area defined for motion detection as just my yard, excluding the street, but since a car with headlights On is very wide, the height is also higher since the width is diagonal across the scene, thus the "bottom" of the object was appearing inside the defined area. I tried to have the object 'IN' instead of "ON" the area, but that then missed people walking through the yard, close to the camera making them much taller in the scene and thus moving the center much higher and out of the defined area. By tipping the camera to the side, the street now is horizontal in the scene and the long car + headlights object is now much less tall, so the bottom of the object does not enter the defined area.

Experimenting with each scene and each of the possible rule features have eliminated a lot of false detections, but I still have to periodically alter the size or shape of an area or the location of a line to take into account the changing lighting conditions of the scene as the seasons change.

But, defining a smaller area for the rule or using "Crossing a line" have pretty much eliminated the problem of lightning causing false detections.

[ryan]

Please also make sure you're running the latest version of the application that was released last week, 4.0.2, as there were several updates to the person recognition algorithm that greatly reduce false positives in many scenarios.