1 image per alert. Right now we are not ingesting short clips because we wanted to minimize response time. As we further test the performance of handling small clips, we may incorporate into service or we may decide going on-prem is the better way forward. TBD.
4.8.5 - BIG NEWS No movement trigger and New AI integration/update
- Thread starter fenderman
- Start date
Changing my first answer (below) after re-reading question. If the first alert does not detect human activity, we still analyze the second frame. We skip alerts when we recognize the "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".
Agreed. We are continuously improving the logic. Once 5G and direct video analysis happens, a lot of current limitations will go away.
Last edited:
- Mar 9, 2014
- 36,892
- 21,407
You will find that local detection will work and sell better. The biggest hurdle for local detection is processing power.
Agreed. Going on-prem is a priority. Hardware is improving quickly and becoming more cost-effective to run AI.
For closed systems like mine, using only a VPN and no open ports, will this still work?
pmcross
Pulling my weight
I noticed that when a text is sent from Sentry that two texts are sent with 1 image. Are there any plans to consolidate to just 1 text message per alert?
Sent from my iPhone using Tapatalk
Sent from my iPhone using Tapatalk
Most of them do not have the blue highlighting but some do have the highlighting in contact with the human.
I am confused though as the demo is showing the system detect a human with just hands coming through a door.
Can the system detect a human with just a portion of a body or not?
edit to correct spelling and grammar.
We continuously train the computer vision model to get smarter. Regarding false negatives, i.e. humans were in the scene but not detected, our testing shows a 95%+ accuracy rate, i.e. we almost always detect humans when they are visible somewhere in the camera. This includes images with partial views of a human, body parts as well as images where the person is not even facing the camera, e.g. walking away as shown in other sample videos on the website.
As we move into processing video streams instead of single frames, the probability of missing people further reduces.
Yes! We received this feedback from other customers as well. Keep in mind, with the Blue Iris integration, text alerts will stop and you will receive alerts through the Blue Iris alerting system - mobile app, website or desktop server.
pmcross
Pulling my weight
That is great news! I would much rather receive push alerts via Blue Iris than text. Definitely looking forward to the go-live.
Have to give some kudos, I was impressed that it got this human.
pmcross
Pulling my weight
So far Sentry has been very accurate. I am running it side by side with Sighthound Video for comparison and so far I’ve received no false positives with Sentry. The experience has been very positive. I’m curious to see how it performs when Sighthound sends a false positive.
Sent from my iPhone using Tapatalk
Sent from my iPhone using Tapatalk
I have gotten some false positives, not many, about 7 out of 800 so far. This is the latest one. Thinks the front seat of the car is a human. Unless someone is hiding in the back seat...
We feel like we have a good handle on daytime images. A lot of our R&D effort now is on making night time surveillance more and more accurate.
-- Sam from Sentry
varghesesa, cool to see your company engaging with people here. Is there anything you could share regarding the security of user images from your end (from areas within your control)? Related question: do Sentry employees have access to the images being processed, and if so, any plans on offering a way to opt-in or out on that?
Great question. All images sent to Sentry are fully encrypted. Sentry employees cannot view or access the images without your consent. When you raise a customer support request that needs us to view the image, we will ask for your permission before decrypting the image.
Sentry AI learns from your feedback. When we launch, you will be able to opt-in to allow us to access images, especially the outdoor ones, for learning purposes. You will always have the ability to opt-out.
Sentry AI learns from your feedback. When we launch, you will be able to opt-in to allow us to access images, especially the outdoor ones, for learning purposes. You will always have the ability to opt-out.
pmcross
Pulling my weight
This is definitely good to know. Are there plans to have other detection types, such as vehicles?
To be honest, we already have a computer vision model to detect vehicles.
We use it for aerial surveillance use cases for drones. Would be interested in knowing what you had in mind with Blue Iris.
pmcross
Pulling my weight
To be honest, we already have a computer vision model to detect vehicles.
That is great news. I’d be interested to know if a car pulled into my driveway, and receive a notification of vehicle make/model/color, if that is possible.
Sent from my iPhone using Tapatalk
lulu5kamz
Known around here
- Sep 14, 2015
- 1,167
- 3,831
