Discussion in 'Sentry' started by sorka, Mar 16, 2019.
Any thoughts on adding this? Are you using your own engine or something like Rekognition on AWS?
It's their own ...if the were using
Amazon it would be more accurate and they would be pointless.
There are thousands of business selling their own machine vision services using Rekognition as the back end.
Rekognition is not a consumer oriented front end. They cater to business that want to provide services just like what Sentry is doing but with productization and integration with 3rd parties.
Again, sentry would be useless as the BI developer could do the same via rekognition. They are not providing any front end. Furthermore they would have no control of the accuracy as they claim to be making improvements. Rekognition is miles ahead of anything sentry is doing.
I suspect your right. I've been using Rekognition with video segments and it's pretty darn good at doing hierarchical classification of objects and returning that as a nested json response but I go over the free tier every month with sending clips from two cameras.
How much would it cost if you paid?
I already do pay but it's a nominal amount each month and it varies.
So what's a "nominal" amount and for how many clips/occurrences?
The first 1000 minutes of video are free. After that it's 12 cents per minute. Sometimes I'm just under and sometimes I'm over by a couple hundred minutes so the cost can vary a lot but so far hasn't been killer. I started out using images but discovered video offered way better results.
if you can code with C#, node or python, you can use DeepQuest AI on your own server for free and without sending your cameras images to 3rd parties. It offers several features, vehicle detection among them. Imho it is pretty damn good, especially considering that it is free.
Also interesting regarding vidoe analysis:
All our tech is in-house.
-- Sam from Sentry
How are you submitting video to Rekognition? I’m curious about testing it out. I didn’t know if you coded something to send video from BI, or if it was as simple as sending video clips via email?
Any update on Vehicle Detection? This is a very important part for anyone that has a camera facing a driveway, street or anything else. Seems that a car pulls up motion takes a snapshots and is sent to sentry and it is rejected as non human then a person gets out of the car and walks up to the house and it's never detected.
In your scenario, I believe the underlying issue is to capture more frames per clip than just one, which we plan to do. When we are able to do so, regardless of car or no car, we would provide more accurate determination of human activity within the clip.
We have vehicle detection already. Other use cases from customers have shown it is needed.
The roadmap for the next release has not been determined yet.
-- Sam from Sentry
I do think more frames would also be ideal for several reasons, but I want to know when a car pulls in my driveway even if no one gets out.
More frames would probley help with the times that sentry fails to detect a Human in a single frame. 2 or 3 frames would double and triple my chances of Sentry to recognize there is actually a person.
Any update on vehicle detection? I have missed reports of people pulling in the driveway then getting out and walking up to the house.
Sometime after BI 5.0 release.
Separate names with a comma.