Face and person detection with Deepstack - local and free!

Share your Projects!
103

Anyone using Blue Iris with this? I could do like faspina has done, Face and person detection with Deepstack - local and free!

104

Yes It will.

105

When wrong objects are detected, the probability score most of the time is below 0.5 (0.4 is the default benchmark). You can increase this benchmark to effectively filter wrong detections.

106

Not sure the component can change this setting yet

1 Like
107

@Klagio use restart = always. Here’s my docker command that I use:

sudo docker run --name=deepstack --restart=always -e VISION-DETECTION=True -v /docker/containers/deepstack:/datastore -v /etc/localtime:/etc/localtime:ro -p 5000:5000 -d deepquestai/deepstack

2 Likes
108

I have been playing with this all week. Here is my wish list

  • Be able to adjust the probability score. This was mentioned in a post above
  • See what Image is being process and have the Object highlighted so that they can be texted or displayed in the front end.
  • be able to bring out a list of objects. For example I want to know how many car or people in my driveway. I know I can do this with a template sensor, its not a big deal.
2 Likes
109

It would be interesting to see the actual image and figure out what it detected as an elephant.

1 Like
110

Yes! I swear I do not have an elephant in my garden.
Is it possible to see the image (portion) that caused thar recognition?

1 Like
111

The camera sees what the eye misses.

The elephant in the room with machine detection is false positives :slight_smile:

112

I wish to have the state hystory, like we have with sensor, is this already there (and I can’t see the obvious) or is it missing?

113

@OlafenwaMoses
Could it be possible to detect a person when is recognized the back differently from the front? In Hoime Assistant we could use that as entering and exiting ( a room for example)

114

I am pretty sure that HA keeps the state history of every sensor, unless you exclude it from the history component.

115

When I click on it it (image_processing.xxxxxx) says
No state history found

116

Hi, if I want to detect person AND dog, do I need two of the below configuration (one per person and one per dog)? Or I can put two targets?

image_processing:
  - platform: deepstack_object
    ip_address: localhost
    port: 5000
    scan_interval: 20000
    target: person
    source:
      - entity_id: camera.local_file
        name: person_detector
117

@robmarkcole @masterkenobi
By your experience is 320x240 enough as resolution for person detection (no need for face)?

My camera has 3 settings 1920x1080, 1280x720 and 320x240, so I wish to use little bandwitch and I think also image processing is less heavy by using the lowest definition

118

It did not work with me, I have the configuration below, with a cam local_file with this img, and face_counter shows 0
`image_processing:

  • platform: deepstack_face
    ip_address: localhost
    port: 5000
    scan_interval: 10000
    source:
    • entity_id: camera.cam_quarto
      name: face_counter`
119

@OlafenwaMoses are images downsampled by the model?

1 Like
120

did you call the service manually? I am noit sure it does the first scan when HASS starts

121

@Marcos_Felipe I have the same problem. When restarting HA the entity shows 0 faces. When I train a new face and hit image processing scan the 0 turns to unknown. Regardless of the taught faces it doesn’t even detect the amount of people within the local_file image.

122

yes, I checked and the service is active, when it does not start the status = unknown