[APP][Pro] FACE++ Face Recognition for Homey

Apps
1

Face++

Face recognition for Homey

Use snapshots from a camera connected to Homey, and let Face++ do AI based face recognition. For every face the app detects, a flow will be triggered. Homey will automatically try to match the faces to a set of known faces that you can upload yourself. The app will also estimate gender, age, emotion, and whether the person is wearing (sun)glasses or a mask.

Face++

Use cases

  • Presence detection
  • Security
  • Face based access control
  • Fun

Privacy

The Homey app uses a cloud based face recognition service called Face++. Using the Face++ service has severe privacy implications. Use at own risk. Read and understand your local privacy legislation, and the terms of use from Face++.

Performance

The results depend very much on the quality of the snapshots. A face will be best recognized if the lighting is good and when the face is frontal with a width of at least 75 pixels. In practice it means you need to have good control over when the snapshot is taken in a videostream. The free Face++ service allows you to upload around 1 or 2 images per second. The face detection usually is ready within 1 to 3 seconds from uploading an image (via an action flowcard).

Test image

To make testing your flow easier, a test image is available as global image token. It has five faces in it.

Face++
Face++1024×767 98 KB

Example flow

8 Likes
2

FACE RECOGNITION IS HERE!

Screenshot_20200428-171924_Homey
Screenshot_20200428-171924_Homey1080×2220 472 KB

3 Likes
3

Love it. however app not starting after fresh install.

4

Is a bug. Has been fixed in 1.0.1

5

version 1.0.1 is now available on Github. It fixes some small issues.

6

Any idea where this is coming from?

image
image800×438 128 KB

7

What did you do when this happened? And where does it pop-up?

8

By choosing a picture from my iPhone for “adding a new face”. Select the pic from the library and this pops.
App version 1.0.2.
Pic is 312x406, 20KB

9

I think I broke something in v1.0.2. Will look into it. Thx!

1 Like
10

Could you check if an app restart fixes it (after saving the API keys)?

11

Sure.

Restart the app didn’t work.
Needed to disable/enable the app. Didn’t work.
Added api key and secret again, then it worked.

12

Cool. New version v1.0.3 fixes this.

1 Like
13

Working, cheers!

1 Like
14

The app has been released just now in the appstore. Get it here:

1 Like
15

Great work!
Seems to work fine in my first tests

1 Like
16

Geniuos! Works perfect on all cams.

1 Like
17

And what kind of camera’s are you guys using?

18

I have tested it myself on the Ring Pro, a Foscam FI9900P, and a Maginon IPC-1a (aldi). For all camera’s the tricky part is making the snapshot at exactly the right moment.

The setup that works pretty good for me is this:

  • Use my Ring Pro to do motion detection just outside my front door
  • This triggers face detection from a second camera (Aldi Maginon) that is mounted inside, looking through the window right next to my front door
  • I make snapshots at 0, 1, 2, 3 and 5 seconds interval

Afbeelding1
Afbeelding1310×650 33.7 KB

result: { origin: ‘Hal’, label: ‘Robin’, confidence: 81.437, token: ‘xxxxxxxxx’, quality: 69.116, gender: ‘Male’, age: 40, emotion: ‘neutral’, normalglass: true, sunglass: false, mask: false }

1 Like
19

I’m using UniFi camera’s with the Protect app. You can use almost any camera which has a public JPG snap function. Just create a virtual ip camera device and point the url to its public JPG

2 Likes
20

What I did was create 3 flows.

1st - When -> Motion sensor is triggered -> Enable flow 2 -> Disable flow 2 after 2 minutes
2nd - When -> Every 2 seconds -> Try to find face from camera
3rd - When -> Face detected -> Run wanted actions and disable flow 2

This way you give the detection a starting and end point. It will try for a maximum of 2 minutes to find a face. Seems to work fine.

4 Likes