Person and face detection using image_processing with Sighthound

I will trust your config is correct, any log errors?

No, I dont see any errors. No need for a reply, I will keep troubleshooting. Just thought something obvious would be pointed out. Thanks!

Updateā€¦
Now its working, I did reset a few times before, I added dlib_face_detect and reset again and now it looks like they both work. Thanks again!

1 Like

@robmarkcole

Robin, would it be possible for you to add a car detection functionality to this component (e.g. state_display: cars, image_processing.detect_cars event)?

I have tried ALPR, but the way it is configured, it counts vehicles only if it detects a license plateā€¦
I have tried the Sighthound API vehicle recognition manually and it is amazing!

@robmarkcole

Hi Robin,
I would also love it if you are able to add Sighthounds car detection to this component.

Thanks,
Hans Magnus

Hi guys, have you tried out the Tensorflw component with a cars model?

@robmarkcole Noā€¦ I thought it would have to be trainedā€¦ Can you point us to this model?

Iā€™ve not looked in depth myself but you could checkout https://github.com/ahmetozlu/vehicle_counting_tensorflow

@robmarkcole
Iā€™ve just setup car recognition with a generic tensorflow solution as described in your articles. I have used faster_rcnn_inception_v2_coco model and it works just fine for my purposes.
I am impressed with the quality of the walkthrough you have put together!
Thank you so much!

1 Like

Does this recognice who is on the picture?

Hi Rob. Nice work, just what Iā€™m looking for.
I am just about to install the component, and was wondering if there is a way to only send a request to process an image when required. I use the motion detection sensor of my cameras to save a .jpg image, and then display that image in HA as the last known motion. I would like to get Sighthound to process this image only when it is updated.

In my config file I have created a virtual camera as follows:

# Camera device from still image
camera:
  - platform: local_file
    name: camera_courtyard
    file_path: /config/www/images/courtyard.jpg

I would like to specify camera.camera_courtyard as the source for Sighthound, but to request the image to be processed only when motion detection is triggered. As it stands now it looks as though you can only specify a fixed interval for the scan_interval.

Currently your only option is to set a long scan_interval then use the scan service triggered manually or from an automation.

Thank you for responding. Having had a quick look at your code it looks like you control when the image is sent to Sighthound. Would it then be possible to add functionality to be able to specify a.) what image to send and when to send it (from within an automation).
Ideally with multiple cameras I would like to select which camera image to send for evaluation, and when to send it.
Is there any chance of adding these options in the future?

This request came up a couple of time but needs to be implemented at the camera platform level

Do you have updated instructions for loading this custom component? Iā€™m getting the following error in the logs:

Integration sighthound not found when trying to verify its image_processing platform.

I have the contents of the file sighthound.py in custom_components/sighthound/image_processing.py

I had it working earlier but something broke when I upgraded to the latest HA 0.92.1 today. Wondering if anything changed in the way the custom component needs to be loaded.

Thanks!

I have the same error after upgrade to 0.92.1.

All please update to release v0.2. Thanks

Thanks! that works

1 Like

By far the easiest setup of all the human detection camera services. Just a pity about the limited API called.
Thanks Rob.

Hi Rob
From your description It looks like you have recently changed from a fixed scan interval to needing to call the image_processing.scan service.

Could you give me an example of how to call the service in an automation?
Many thanks.

Sorry, just seen the example.