MR60FDA2 - two issues: 1) not reliably detecting fall and 2) device says that a still person is not present

Hi - I am evaluating the MR60FDA2 using Home Assistant. It took some doing (esphome didn’t initially like the C3), but I was able to get it up and running on HA. I am experiencing a couple of issues:

  1. presence detection doesn’t seem to work well. I can be sitting at my desk typing within range and moving, and it will say that a person is present. If I am still for a few seconds, it will say that no person is detected. If I move a little, it will say that there is a person present again.

  2. I can’t get it to reliably register a fall (except it has done a few times randomly). But if I simulate a fall more or less directly under the device, it does not register the fall. This is my main requirement. I have played with the various sensitivities and they seem to make no difference.

Anyone else have any experience or ideas?

Thanks.

Hi there,
And welcome here,
So can you post a picture? probably not yet but soon, So what method of power?
I have had to go through several USB chargers to find a quiet one. batteries work ok but not for this imo.

BTW this unit has a C6 on it yes/no ?
HTH
GL :slight_smile: PJ :v:

EDIT: Apparently I can’t embed pics in the post so I put them on DropBox. The link is: here

Hi PJ -

Thanks for your response. Yes, it is a C6. My mistake.

I have attached a screenshot of the sensor history from HA. No one was in the room after 18:00. We were in bed (upstairs) by 21:30. You can see that there are a lot of spurious fall detection events between 18:00 and 12:30, but strangely none after that.

I have also attached a screenshot of the settings in HA. Note that I am sitting under the device, and it shows no presence.

Finally, I have attached a pic of how it is mounted. It is powered by the USB cable from a wall socket that has a built in USB charger port. I will try a separate power block. The unit next to it has no power connected, so can’t be interfering.

Any help would be appreciated.

Thanks,
John

1 Like

Hi there,
Nice work. :+1: I see those pics, Looks like you have it installed correctly.
I’m curious two things , Is the Light on always? have you measured the heat it generates?
I know your probably LOL at that but , I’m hmmm… Something is generating waves.
turn it 90 degrees and also test that.
These things are almost an Art form in getting to work reliably and repeatable. Takes some fiddling and work. by no means plug and play IMO.
Your on the hunt so you’ll get it…
A) power supply
B) Temperatures ambient (try lowering the temp)
C) Orientation
D) TEST
is the check list.

HTH
GL :slight_smile: PJ :v:

Thanks for the tips.

I have not had the LED on. It’s kinda neat to play with, but I am not convinced that there is a lot of practical value in that feature.

  • I have replaced the power supply with a (hopefully) clean power block and will test some more.
  • It’s hard to change the ambient temp, and in any case, I need it to work at least up to ~25C…
  • I will check the orientation if the power doesn’t help.

Do you know what the Height Threshold and Sensitivity control exactly? I have looked in the code, and the Sensitivity numbers are translated to larger numbers, e.g. {1, 2, 3} = {3, 15, 30}, but not sure exactly what these represent… number or readings/frames to average over? Higher number is more sensitive?

FYI, the other unit is a C1001 from DFRobot. I built a custom component for HA for it, but it seems to be quite unreliable. Maybe as I gain more experience with this unit, that one will make more sense…

Thanks again.

1 Like

Hi there,
Yep , it’s cutting edge stuff.(mmWave) Maybe support can comment on the nature of the Numbers in the code , it say’s meters in the docs that I have seen. but they haven’t put out much on the code side. Everyone’s guarding their stuff so-far. Stay tuned and keep testing.

GL :slight_smile: PJ :v:

Hello, we are glad that you are interested in our products. Thank you for your feedback, I can give you some suggestions here. Currently, for the MR60 model, we have submitted a PR for both radars to ESPHome pending their merger. So you may not be able to find a more detailed description and guidance about this radar inside the public documentation for the time being, but I’m sure it won’t take long and you can check the preview version of the introductory documentation first at the link below.

1 Like

The main function of fall radar is to detect falls and the main factor that affects how it works is height. There are two thresholds for this, one is the height at which the radar is mounted as well as the threshold height at which the fall detection is triggered. Currently, the radar only supports top mounting (that is, mounting in the ceiling position of a room). Based on your description, it looks like you are planning to test side-mounted, which we think could be one of the main reasons for false recognition of falls. Another reason is that you may not have set the parameters for the two heights of the radar correctly. You may need to take a little time to mount the radar on the ceiling and measure the mounting height to set the correct mounting height for the radar. Then, you need to find out what the approximate height above the ground is when the subject is detected as having fallen, and use that value to set the Height Threshold. that way, you won’t let the radar mistake the act of sitting down for a fall.

1 Like

Hi there,
Awesome, Thanks for the response , However the picture from the op’s dropBox shows it is ceiling mounted next to a LIght and Above the subject. I do believe you have provided some great info on the parameters all along. :+1: The Total Height is One setting, and the Highest point a subject maybe below the sensor ? is that what you mean or as if the subject were collapsed/fallen on the ground in a mass the highest point of that mass?
ie. a 9’ foot ceiling, and the 2.5 feet from floor if he lays on his side. Ah’ this makes sense with the last statement, thank you " that way, you won’t let the radar mistake the act of sitting down for a fall "
Let see what he(op)figures out, I don’t have one but should get one it looks put together well. I like it’s a C6 as well.
GL :slight_smile: PJ :v:

Thank you. I already have the device integrated with HA, but it wasn’t as plug-and-play as advertised. That was no big deal (for me). I have also been through all the available documentation several times. What is not made clear in the documentation (as PJ mentions below) is what exactly the Threshold Height means, which you clarify somewhat in your next post. Also, some clarity on the Sensitivity means/how it works would be useful.

My device is mounted on the ceiling. I measured the Install Height and set that at 2.6m. I have played with Height Threshold, at around 0.4-0.6m, and I get false fall indications when no one is in the room at all. This morning, for example, there were several over a period of about 15 min (see pic in this location).

I am pretty sure that these false positives were caused by someone moving in the room above where the device is mounted. If so, this is way too sensitive and the device needs to be able to screen for this kind of thing.

Thanks!

Again this morning, after I put the sensitivity down to 1, the device is giving false positives in response to someone walking in the room above.

Can someone please help me try to figure this out? Or is this just not something that I can fix? If not, the usefulness of this product will be very limited.

Thanks.

Hi there,
So I haven’t received this one yet to test with but I would offer this. The millimeter wave radars are a new area in IOT you can check everywhere it’s a dark ART to get it perfect and the environment has to be pretty close too. SO that being said.
If the mount it is connected to is moving vertically ? even a millimeter and if the sensitivity is set very high it will trigger makes sense right?

I have a demo on here of the discrete board connected to a xiao directly and the accuracy was very high albeit the environmental requirement were specific to get repeatability. YMMV

You may need to add some kind of dampening material under it or move it closed to the adjacent wall. I had a similar situation with testing and required a mini Gantry (see pic)

HTH
GL :slight_smile: PJ :v:

Thank you for your feedback, we will recheck the current firmware for the situation you described. We will see if we can reproduce your problem, we will reply you here as soon as we have the result, please wait. We apologise for any inconvenience caused.

Our product manager mentioned a key problem point. Is the next door above your radar mounting location very thin? Does it make the radar jittery when you walk around? If so, maybe this is the main reason why there are false messages being reported.

Hi - Thanks for your reply. I am pretty sure that this is indeed the issue. The problem is that the house is built to normal American standards, and it does not move a lot. I would estimate maybe on the order of 1mm deflection when someone walks above. The unit should not be so sensitive that this would give a false positive.

thanks,
John

Thanks. I am sure that this minor movement is the issue, but I have the unit set to sensitivity = 1, which I think is the minimum. There should be some way to filter out this kind of noise, as it is at least two orders of magnitude less that that which would trigger a true positive.

Thanks,
John

1 Like

Hi again -

I have been gone for a few days, which gave me a good chance to look at how the unit performs. My house is set to Sensitivity = 1. It is clear that walking around above it makes it give false positive readings for falling. This is in a house that is built to a high standard for US houses, i.e. it’s not masonry, but it is a wood frame, so a small amount of deflection is to be expected.

I think this device has a lot of potential, but these false positive readings ruin it. I would be happy to collaborate on a way to ‘debounce’ those readings. Are you willing to share more information on the API so that I can access the actual data stream and/or possibly develop an algorithm to remove the false positives?

Thanks,
John