Google Vision Plugin

Posted on
Fri Sep 29, 2017 8:03 am
vtmikel offline
Posts: 628
Joined: Aug 31, 2012
Location: Boston, MA

Re: Google Vision Plugin

Version 1.0.3 has been released with bug fixes and support for Logo Detection.

My initial tests didn't come out well for the Logo detection, so I'm interested if anyone finds success with it. As an example, the logo detection isn't able to detect USPS logo in this picture that I found using Google Image Search: https://i.ytimg.com/vi/e67_3l3N9hQ/maxresdefault.jpg

Thanks

Posted on
Tue Oct 03, 2017 2:15 pm
vtmikel offline
Posts: 628
Joined: Aug 31, 2012
Location: Boston, MA

Re: Google Vision Plugin

I'm interested in hearing the experiences of anyone who has implemented the plugin. I've found that Google's API doesn't produce useful labels for when a person is found in a image. I took some manual shots of myself in my own cameras in my entryway and no label for a "person" appeared. The Face API does not respond unless an actual face can be seen in the image, meaning that the person is staring at the camera, which doesn't work well for security cameras that are usually placed above.

Objects, seem to be identified well. I don't have my outdoor cameras installed yet, but based on my experience with my Garage camera, I imagine that recognizing a delivery vehicle will be doable with this service.

Mike

Posted on
Tue Oct 03, 2017 3:37 pm
RogueProeliator offline
User avatar
Posts: 2501
Joined: Nov 13, 2012
Location: Baton Rouge, LA

Re: Google Vision Plugin

I'm interested in hearing the experiences of anyone who has implemented the plugin.

I want to give this a shot, but my security DVR died a couple of days ago... I'm about to throw in the towel and do an extended warranty credit card claim, so hopefully can get it replaced before too long. I do have a couple of IP cameras I could just test with directly, so if I get some free time I may -- but can't grab a still for, say, a delivery driver off a recorded video.

I'll let you once I get some data/results.

Posted on
Mon Oct 09, 2017 2:10 pm
vtmikel offline
Posts: 628
Joined: Aug 31, 2012
Location: Boston, MA

Re: Google Vision Plugin

I've merged this with a new plugin: viewtopic.php?f=134&t=19192

Posted on
Wed Feb 07, 2018 4:37 pm
Swancoat offline
Posts: 503
Joined: Nov 20, 2009
Location: Houston

Re: Google Vision Plugin

This thing is cool. I've got it checking to see if a package has been left at the front door.

The plugin works great, the only problem is that Google Vision never identifies a package. It's a pretty boring security cam shot of my doorway, and Google returns stuff like 'wall', 'structure', 'building' etc... (lots of brick in the shot which I assume drives a lot of that).

Too bad we can't train the api a bit (Like I could upload pics and just add labels of what I see, until it recognizes 'box' or 'package').

Anyway, they say it learns and gets better, so wait and see I guess.

Plugin is killer though.

http://nerdhome.jimdo.com

Posted on
Thu Feb 08, 2018 7:45 am
vtmikel offline
Posts: 628
Joined: Aug 31, 2012
Location: Boston, MA

Re: Google Vision Plugin

Thanks for sharing your thoughts. You might be the only active user other than myself. I bet everyone's experience with this plugin is going to be quite different.

Some random thoughts and questions from me:

  • What is your camera setup? I'm interested to hear how you are getting your still image from the camera and making it accessible to Indigo on the local drive. Are you using Security Spy like me?
  • How are you dealing with getting multiple results from the Image plugin over time? I haven't implemented anything yet, but I eventually want to make it so that: 1) Camera detections motion 2) Image plugin triggers, sending 10 images over 20 seconds. 3) If >80% of the results come back matching X condition, trigger an event.
  • I'm personally going to be installing 3-4 additional outdoor cameras in the next 6 months or so, and will be increasing my testing with this plugin.
  • I really wish google or Amazon supported specifying where in the image it sees a object. I recently bought a second car, and made enough room in my garage to fit the second car. This now makes my "car is home" variable totally worthless, because the image plugin cant tell me which car it sees. I'm going to have to put a separate camera on each car.
  • Amazon does allow training of faces using their services. I haven't tried it yet.
  • My impression so far with the Amazon and Google services is that they are intended for consumer camera pictures. Big faces. Pictures with a single object in it. I found detecting a person on a image to be difficult unless they are staring at the camera. Etc. When you take a picture with a wide range and a lot of things going on, the results are inconsistent, which is disappointing. I had little success getting the services to tell me it sees a delivery truck (never mind a package).

P.S. I saw on your blog that you do video matrix and distribution stuff. I did this recently for my new house setup, using Just Add Power. I also built a Indigo plugin for it.

Posted on
Thu Feb 08, 2018 9:29 am
Swancoat offline
Posts: 503
Joined: Nov 20, 2009
Location: Houston

Re: Google Vision Plugin

My camera setup is basically IP cameras fed to Security Spy like yourself.

So the whole process of putting images into a directory and then uploading was a bit clunky, and I'm not quite sure I'm doing it the best way yet.

Basically, for my front door camera, it is recording video on motion and then continuously grabbing a still every 60s and overwriting the previous still. (I'm at work right now, and don't have Security Spy in front of me, so I can't remember the exact details on why I set it up like that. I think I wanted to have a statically named file, and that could only be done with the 'continuous' method as opposed to just when motion was detected).

Another wrinkle I had was that Security Spy is not running on my Indigo mac, but rather an older iMac I have that basically just runs Security Spy. So I think I'm using Security Spy's 'webcam' method to put the jpg on my Indigo mac via ssh. That seems to be working fairly well.

So now using the Cynical Security Spy plugin to trigger on motion detected on that Camera, I use your plugin to upload the image (and then do it again, delayed by 60s to make sure I get an image of just the package, and not the delivery guy).

I may want to change the timing so that a new still is being recorded every 10 or 15s and then narrow that interval for when the second image gets uploaded. Maybe even upload a third.

As for the trigger, I currently get an email whenever there is motion on the camera, and if it is a delivery, I can either reply to the email with a pre-determined subject line, or alternately I have an action I can just trigger manually from the Indigo app on my phone. Those simply set a flag variable that I have a package waiting, and when the next person comes home it makes an announcement on the whole house audio that there is a package at the front door (since we never actually come in there, and usually just come in through the garage). I'd like to have the plugin handle the setting of that flag!

http://nerdhome.jimdo.com

Posted on
Thu Feb 08, 2018 4:20 pm
virgahyatt offline
Posts: 132
Joined: Jan 11, 2014

Re: Google Vision Plugin

How many API calls do you guys average a month? I have been looking at the Sighthound API (https://www.sighthound.com/products/cloud) for image recognition. However, if you need more than 5000 calls a month you have to pay $49/mo. But if it lives up to their claims it's quite powerful.

Posted on
Fri Feb 09, 2018 10:22 pm
GlennNZ offline
User avatar
Posts: 1555
Joined: Dec 07, 2014
Location: Central Coast, Australia

Google Vision Plugin

Hi

Thanks for this plugin (the Image one replacing above) it looks very interesting. Have a large number of outdoor cameras and was running OpenAlpr for a while (and had put together a plugin for its service). This is a Licence plate reporting service - was quite cpu heavy and wasn’t impressed with results (daylight okay, night not so). The OpenAplr sites report better success - but they also moved to a charge service for the better updated options, so eventually wasn’t interested. An upload service like these is probably better option - although does add a bit of delay...

These APIs and your plugin look like a great replacement and at least a fun way to kill some time! Thanks again, I might add that the sight hound API also looks like an interesting addition.....

Glenn

Posted on
Sat Feb 10, 2018 3:50 pm
vtmikel offline
Posts: 628
Joined: Aug 31, 2012
Location: Boston, MA

Re: Google Vision Plugin

I enable a schedule that sends images every 20 seconds, based on motion from my camera. I usually keep that schedule enabled for 3 minutes after a motion event. My Google Cloud dashboard shows 3000 API calls in the past 30 days. This is for a single camera. I'm still riding their free credits, which will end in September 2018. Looks like if I were paying, I would be averaging $3 a month.

I can look into adding support for the Sight Hound API. Can you all take a few still shots from their security camera and send them to Sight Hound and let me know if you see significant improvements over the Google and Amazon APIs? I'll do the same.

Posted on
Mon Jan 21, 2019 9:52 am
boekweg offline
Posts: 70
Joined: Oct 02, 2010
Location: Netherlands

Re: Google Vision Plugin

I have a question. I installed the plugin and have the system up-and-running in google cloud vision. OCR works well, it triggers on my license plate of my car and I can use that to get access to my garage f.i.
Is it possible to not only detect faces but also add a name to that face, i.e. i can unlock my door by looking at the camera, or sent a message that a known person is at the door?

Great plugin ! Thanks a lot

Posted on
Fri Jan 25, 2019 6:24 pm
vtmikel offline
Posts: 628
Joined: Aug 31, 2012
Location: Boston, MA

Re: Google Vision Plugin

boekweg wrote:
I have a question. I installed the plugin and have the system up-and-running in google cloud vision. OCR works well, it triggers on my license plate of my car and I can use that to get access to my garage f.i.
Is it possible to not only detect faces but also add a name to that face, i.e. i can unlock my door by looking at the camera, or sent a message that a known person is at the door?

Great plugin ! Thanks a lot


I started to respond to this and never sent. Apologies for the delay.

Amazon Rekognition does have this feature. I'd be happy to build support in for it if you found it useful, but I have my doubts, so I think we should test first.

The way that facial recognition works is through your private AWS account, by creating a image collection.

Here is the info:
https://docs.aws.amazon.com/rekognition ... tions.html
https://docs.aws.amazon.com/rekognition ... ction.html
https://docs.aws.amazon.com/rekognition ... Image.html

You are supposed to create your own collection of faces. Then, subsequent images that are sent for analysis can be compared against your private collection.

I'd suggest trying this manually, and seeing if you can successfully detect and recognize faces with your security camera images. My fear is that this feature of Amazon is ideally used for SLR or phone images, where the faces are a large part of the image, and at eye-level.

Posted on
Tue Jun 04, 2019 1:11 pm
boekweg offline
Posts: 70
Joined: Oct 02, 2010
Location: Netherlands

Re: Google Vision Plugin

Thanks a lot, I will give it a go. In the mean time I have the netatmo welxome up and running, this allows you to add namens to faces. The Indigo Plugin for this works well and enables me to trigger on recognized faces.


Verzonden vanaf mijn iPhone met Tapatalk

Who is online

Users browsing this forum: No registered users and 0 guests