Understanding/preventing motion false positives outdoors

[lite1]

9 motion detected videos @ 10 to 17 minutes apart from 1 camera mounted about 4' off the ground and pointed slightly downward with motion sensitivity @ 80%. In looking at, at least the first 10 seconds of each video I can not see anything that would trigger motion detection. The camera axis aims toward the SE and at this time of morning the sun is transiting across the upper left quadrant of field of view.   

Since I am only on Basic/free subscription, I expect recommendation would be: 1) Reduce sensitivity; 2) aim camera more downward; 3) move camera to a different position. I expect that if I had a paid subscription that using Activity Zone would likely reduce false positives. I've been trying to avoid having to pay a monthly fee, but might conclude that that is the only or easiest way to have things work better.

I had just recently moved this camera closer to the cat door that I want to (but not exclusively) monitor; and with prior location (which was higher vertically and farther away even with sensitivity at 90% cat activity at almost right angles to axis of camera view was being missed. While I am fine in the new position of camera to experiment with reducing sensitvity, the current field of view is ideal to "see" both cat door and approach to garage so am reluctant to relocate. While I want to solve this problem, I am more interested in principles of camera placement to minimize false motion positives especially those that are due to changes in sunlight/shadow. I have evergreen trees so even when windy I am not asking Pro 2 to deal with excluding triggering on leaf movement.

It seems that for some locations and environmental conditions that the sweet spot for camera location, angle, and sensitivity, might be fairly narrow, and if so this makes use of system more challenging. Are their seasonal changes and/or time of day changes that lessen or increase the same camera and its false positives? Since this IR motion detection, it'd seem to me that there might be differences between snow on the ground versus a hot summer day.

Thanks for any insights and general tips. 

[StephenB · Accepted solution]

---

@lite1 wrote:

9 motion detected videos @ 10 to 17 minutes apart from 1 camera mounted about 4' off the ground and pointed slightly downward with motion sensitivity @ 80%. In looking at, at least the first 10 seconds of each video I can not see anything that would trigger motion detection. The camera axis aims toward the SE and at this time of morning the sun is transiting across the upper left quadrant of field of view.   

Shadows can trigger the motion detection, and that includes clouds that suddenly block the sun/shading the area.

Wind can also trigger it, as the moving air is at a different temp (and carries dust).

Activity zones help, but of course that's only the case when the trigger is outside the zone. 

I've managed to get a reasonable false positive rate for most of my cameras, but there are two that remain troublesome.  One is in the back yard, and the issue there is shadows from nearby trees.  The other is positioned on my front porch - the windows (including the ones in the door) reflect moving vehicles on the street, and reflections of larger vehicles will often trigger that camera.  

FWIW, it'd be good if Arlo had a smart classification for shadows, and maybe also if they distinguished "no substantial motion" from "all other motion".

[lite1]

@StephenB As always thanks for proving helpful and very clear information. Somewhat related to all of this: With a paid subscription and what I think Arlo refers to as their AI ... Is it true that the Arlo cloud/central processing "knows" what area(s) of the camera view led to detecting motion and thus sending a notification? If this is true, would it not be possible for Arlo to overlay an arrow (or triangle) to the area(s) in the video recording that triggered the event, and have such indicator be user enabled/disabled; and that such indicator only be displayed for the first 3 seconds of the recording so that it then is less likely to obscure important information that user wants to see. Has such a feature request been put in the Ideas sub-forum?

From what you have written, I realize again that with fuller knowledge, I have been making an error in reviewing motion triggered videos. Namely that if motion is not readily apparent to me that I then scan the edges of the video frame for anything that has just moved into the frame, and if I can not see anything moving that the video is a false positive. I realize now that while this might be a good place to start analyzing as movement at the edge of the frame is most likely to be at a steep angle to the camera viewing axis; it is definitely not the only movement that will trigger things. It seems that particularly shadow; or wind/dust as the motion trigger could easily start somewhere other than the frame edge, yet still not be obvious when reviewing a video. Your suggestion of other classifications, I think is excellent and would add to the "I" in their AI, which it seems at this point isn't all that smart.

 when window glass reflections trigger your porch camera, are such reflections actually visible in the video? or did your detective work uncover this as trigger from observing what is going on because you had been in the area when a notification was sent?

As an aside and an unexpected benefit of making lemonade ... the frustrating Arlo behavior gave me a good opportunity to teach the 14-year old neighbor who does some gardening with me about false positive and false negatives - a topic that it seems is not taught in school.

[StephenB]

---

@lite1 wrote:

 when window glass reflections trigger your porch camera, are such reflections actually visible in the video? or did your detective work uncover this as trigger from observing what is going on because you had been in the area when a notification was sent?

---

The reflections are visible in the video - so no detective work was needed. 

I do have recordings there it isn't obvious.  So it could be dust/wind, or it could be that whatever triggered the PIR sensor didn't actually enter the field of view, or that the recording lag resulting in the object not being shown.

---

@lite1 wrote:

 Is it true that the Arlo cloud/central processing "knows" what area(s) of the camera view led to detecting motion and thus sending a notification? If this is true, would it not be possible for Arlo to overlay an arrow (or triangle) to the area(s) in the video recording that triggered the event, and have such indicator be user enabled/disabled; and that such indicator only be displayed for the first

When you use Cloud Activity Zones, the video is analyzed by the cloud (which knows the activity zone configuration).  It doesn't know what actually triggered the PIR sensor - that isn't really possible.  The sensor just triggers when there is a change in the IR that reaches it.  There isn't any way to tell where that IR came from.

In principle, the cloud could create an overlay of some kind.  There would be some disadvantages if that is hard-encoded into the video itself:

• the overlay might obscure some information in the scene that is quite important
• the overlay would require the original video to be re-encoded, which would reduce the quality.

There are some ways to add a graphic overlay to a video - DVD and BluRay subtitles work that way.  That would allow the overlay to be turned on / off.   Player support for those overlays would be a problem though - generally MP4s would use text subtitles, and not the graphic ones.

[lite1]

@StephenB When you use Cloud Activity Zones, the video is analyzed by the cloud (which knows the activity zone configuration).  It doesn't know what actually triggered the PIR sensor - that isn't really possible.  The sensor just triggers when there is a change in the IR that reaches it.  There isn't any way to tell where that IR came from.

I am confused! It sounds like what you wrote above implies that the "analysis" that motion detection is using is a change in the total amount of IR sensed and whether that changes at a fast enough rate for the system to then decide that "this is due to motion." But if this is true of how things work, it doesn't seem to be compatible with Arlo and forum information that cameras are more sensitive/more likely to trigger when movement/IR changes are in a direction that is at least oblique to the camera viewing axis. Can you uproot my confusion? I expect that your bold statement above is true and that is why a cloud moving across the sun will be a trigger event.

I am going ahead and marking your earlier response as solution, thanks.

Thanks for good explanation on video overlay and example of subtitles. I guess a possible implementation of providing richer information on video would be a subtitle toggle (again only visible for the first 3 seconds) that would indicate which zone had triggered the recording (if indeed Arlo "knows" that and you indicate that they do NOT), meaning that the frame would be considered a clock face so the sub-title might be Z2, 10; or Z2-4 meaning for the first that motion was noted somewhere in the pie wedge going from the center of frame out toward 2 oclcok and similarly to 10 oclcok; and for the 2nd subtitle meaning it occurred in a broader area in the 2-4 oclcok region. Of course this is wishful thinking on my part that Arlo would get there in my lifetime.

[StephenB]

---

@lite1 wrote:

@StephenB When you use Cloud Activity Zones, the video is analyzed by the cloud (which knows the activity zone configuration).  It doesn't know what actually triggered the PIR sensor - that isn't really possible.  The sensor just triggers when there is a change in the IR that reaches it.  There isn't any way to tell where that IR came from.

I am confused! It sounds like what you wrote above implies that the "analysis" that motion detection is using is a change in the total amount of IR sensed and whether that changes at a fast enough rate for the system to then decide that "this is due to motion." But if this is true of how things work, it doesn't seem to be compatible with Arlo and forum information that cameras are more sensitive/more likely to trigger when movement/IR changes are in a direction that is at least oblique to the camera viewing axis. Can you uproot my confusion?

---

Although the PIR sensor elements don't form an image in any sense, there is a region where the PIR is most sensitive (due to the way the sensors are aimed/aligned).  That is called the "detection zone" in the various camera positioning FAQs, for instance: https://kb.arlo.com/37/How-do-I-position-my-Arlo-wire-free-camera

Another aspect is that the sensor itself has two elements organized side by side.  So the sensor can distinguish someone is moving through the zone into the field of view from someone moving through the zone out of the FoV.  But that can be spoofed by clouds or wind.  And as I tried to say, the detection zone of the PIR won't be exactly aligned with the edge of the FoV - especially in the Ultra where FoV is configured.

There is some more info here that you might find helpful: https://learn.adafruit.com/pir-passive-infrared-proximity-motion-sensor/how-pirs-work.  Though the lens description at the bottom is for one they sell ( https://www.adafruit.com/product/189 ). It's quite large (about 1") and I don't think the Arlo cameras are using anything that complicated.

What happens when the sensor is triggered depends on the camera model and whether it is powered or not.  If an Ultra is powered, then the camera itself will analyze the video stream, and look for motion in the zone.  (It will also process and stream the 3-second buffer).  If the Ultra is battery powered, the cloud will analyze the video stream. and look for motion in the zone.  Either way, if there is no motion detected the stream is discarded.

But if there are no activity zones (which is the case for some of my cameras), the stream will always be recorded.  Those recordings sometimes don't show any obvious motion at all - so I can't tell what triggered the camera.  And if there is an activity zone, there is often some slight motion in the zone.  I suspect that motion isn't always what triggered the PIR sensor - that something else did - but the motion was enough for the system to keep the recording.

I guess the net here is that you will end up with some recordings with no obvious trigger. 

[phaje]

---

@StephenB wrote:

---

---

@lite1 wrote:

 Is it true that the Arlo cloud/central processing "knows" what area(s) of the camera view led to detecting motion and thus sending a notification? If this is true, would it not be possible for Arlo to overlay an arrow (or triangle) to the area(s) in the video recording that triggered the event, and have such indicator be user enabled/disabled; and that such indicator only be displayed for the first

When you use Cloud Activity Zones, the video is analyzed by the cloud (which knows the activity zone configuration).  It doesn't know what actually triggered the PIR sensor - that isn't really possible. 

---

I bought an Arlo Pro 3 set of cameras only a few weeks ago.  I think at some point after that the alert notifications were upgraded such that they include a thumbnail of the area in the picture that triggered the alert.

[StephenB]

---

@phaje wrote:

---

@StephenB wrote:

---

---

@lite1 wrote:

 Is it true that the Arlo cloud/central processing "knows" what area(s) of the camera view led to detecting motion and thus sending a notification? If this is true, would it not be possible for Arlo to overlay an arrow (or triangle) to the area(s) in the video recording that triggered the event, and have such indicator be user enabled/disabled; and that such indicator only be displayed for the first

When you use Cloud Activity Zones, the video is analyzed by the cloud (which knows the activity zone configuration).  It doesn't know what actually triggered the PIR sensor - that isn't really possible. 

---

I bought an Arlo Pro 3 set of cameras only a few weeks ago.  I think at some point after that the alert notifications were upgraded such that they include a thumbnail of the area in the picture that triggered the alert.

---

That is analyzing where the motion is in the video.  It isn't showing definitively showing what triggered the PIR sensor - which could be something different.