Discussion in 'What's New' started by fenderman, Feb 7, 2019.
it varies by camera. There is no one fits all solution.
fyi you can start testing sentry using the setup guide provided above. Note there is an error in the manual, their email address email@example.com the manual has "senty" missing the R.
So Blue Iris still has to detect something and become triggered. The image taken after the trigger is emailed to a Cloud system and they then process it and text me if it is human or not.
I got excited at first but now not so much.
And so you're back to reality at this point in time. Gotta hand off your data and hope for the best because the avg system and consumer software options (Windows) don't exist to solve this locally for a normal user with normal HW. Happy to see interest here because before long, we'll be able to process this stuff in-house as long as software/hardware evolves and the DIY demand is there.
Lets see how it plays out.
I was very pleased to read this, untill i saw it uses email for sending the pics. Makes you dependent on 2 services: email and sentry.
Might go the
amazon route then, they have an API
I'm an Amazon customer/investor and I wouldn't. We need local superpowers, and that isn't that far off from the present. Or, you can trust a really large and capable company with endless amounts of money, and end-goals of I don't quite know what, and hope for the best. I say all that and I doubt Amazon would get involved with gov surveillance. WTF am I even talking about
Blue iris having to detect something is not the issue, it is very easy for BI to detect motion. You dont get a text if its not a human or car. The biggest issue is that you have no control over the exact frame that gets sent. A better option - likely more costly would be to send 1fps to get processed continuously. The release note indicates that this is one of many add ons, including more of BI's own detection improvements.
So BI detects, image is uploaded to the cloud for processing. Then if a match (for human activity in this example) is met, a notification is then sent. So how long realistically would that end-to-end process take to justify using the Sentry solution for real-time monitoring? Or maybe it's not meant for real-time?
It's only a couple of seconds... It's going to take a while to smooth out to make it accurate and usable. It's a step in the right direction. Or you can spend some time setting up zone crossing and get highly accurate alerts.
This is what I am concerned with as well. Granted it would be nice to get accurate alerts, but if there is a fair amount of latency involved then a local solution, such as Sighthound Video 6.0 (once released) or another AI integrator with BI would be the best solution. I am testing Sentry on a few of my cameras, I will report findings.
Full disclosure, I work for Sentry.
The integration is much tighter within the software. Email communication is for general users with or without
The Blue Iris app leverages the Sentry API to get and receive the processed image results much faster. Enabling the functionality is just a mouse-click on a checkbox. No setup required. All alerts are sent via the Blue Iris communication channels: mobile app, desktop or browser.
Look forward to everyone's feedback. Hope this reduces false alerts.
This is Sam from Sentry. Appreciate all the interest around the product. FYI, we are planning to launch in a couple of weeks... Feb. 22.
This is Sam from Sentry. Performance may not meet your standards with the email integration. We are getting 3 sec response when no one is identified and up to 6 sec round trip response time when people are identified. Variables like internet connection speed can improve or affect performance. Working on an on premise solution.
This is good news. Can you provide info on how many images get sent to sentry per alert? Will there be a way to constantly send images 1 or 0.5 fps so that sentry will not have to rely on a BI trigger?
Additionally, what will be the pricing structure?
Even though I am not crazy about the cloud functionality, I wanted to give it a fair test.
I used images from four different Hikvision cameras taken through
Blue Iris and sent about 100 images from each camera through the system.
A false positive is an image that came back as containing a human when it really did not.
A false negative is an image that did contain a human and was not detected.
About one third of the images were during reduced lighting. My definition of reduced lighting is street light or better. One of the four cameras stays in color mode all night as their is enough light to not send it into night mode.
During the day about 60% that contained a human were detected as containing a human.
During reduced lighting about 40% of those containing a human were detected.
I had two false positives, one of which is the attached picture with a guys face on the side. It made me laugh and I am not sure if that should count as a false positive.
The green rectangle on the bottom is from Blue Iris and was on the image I sent. The green rectangle on the face is from the Sentry system and was added to the text I received.
The blue iris motion highlighting and rectangles may confuse their algorithm. On the images you sent was there blue highlighting or green rectangles present near the person?
Sam from Sentry: That's a good point. The blue highlighting can affect the image processing. We also noticed a false negative (person was there but not alerted) when part of the body was covered with the blue highlighting.
Great analysis. Also keep in mind we provide additional logic which suppresses frames for a better user experience. We skip alerts when we recognize an "event" has already been detected and alerted upon. For example, if the kids are playing in the backyard, the backyard camera will send an alert the first time the activity is detected. However, if the kids are playing for 30 minutes, you will not receive alerts for the next 30 minutes while the kids continue to play. The software is smart enough to know it is the same "event".
The problem with that is if the first alert is a non-human, the second will not be detected.
Separate names with a comma.