Improved motion detection algorithm (testing required)
Hi All,
We've been working on improving the motion detection algorithm in SecuritySpy and have posted a public beta for users to test. Anyone who would like to test this and report back on their experiences, this would be very helpful to us for refining this before the next update.
The new algorithm should be much more able to ignore movements such as swaying plants, snow and camera noise. So it should produce fewer false-positive recordings, while maintaining or improving the rate of true-positive recordings.
Before testing please reset the motion sensitivity for all cameras back to 50 (the middle of the sensitivity slider), as if you have changed this for the old algorithm your setting may not work well with the new one. 50 should work well in most cases, but please let us know if you find that you have to change this in order to get reliable motion detection (as we can tweak the default based on the feedback we get).
Many thanks in advance to all testers!
We've been working on improving the motion detection algorithm in SecuritySpy and have posted a public beta for users to test. Anyone who would like to test this and report back on their experiences, this would be very helpful to us for refining this before the next update.
The new algorithm should be much more able to ignore movements such as swaying plants, snow and camera noise. So it should produce fewer false-positive recordings, while maintaining or improving the rate of true-positive recordings.
Before testing please reset the motion sensitivity for all cameras back to 50 (the middle of the sensitivity slider), as if you have changed this for the old algorithm your setting may not work well with the new one. 50 should work well in most cases, but please let us know if you find that you have to change this in order to get reliable motion detection (as we can tweak the default based on the feedback we get).
Many thanks in advance to all testers!
Comments
I've just uploaded a new beta (3.4.3b6) that has some further improvements. Specifically, it will detect and filter out whole-frame changes caused by stuff like camera focus and exposure adjustments, and it will detect and filter out rain.
CaliObserver - the new version now does not trigger on the false-positive rain videos you sent. Rain has quite a distinctive pattern of motion (short spikes that are spaced apart) so this is possible to detect and ignore. This should be a significant improvement.
However, a couple of videos you sent feature blobs moving across the image - one appears to be a bug flying past the camera and another is a spider web with an out-of-focus water droplet moving around. Unfortunately these aren't possible to filter out without the risk of missing real motion, because their signature is very similar to movement that you will actually want to capture (people walking, cars passing etc.) The main problem is that it's dark and the camera's IR illumination is amplifying the effect of these small objects moving close to the camera. It's best to periodically clean your cameras of any dirt and debris, as these will understandably impair SecuritySpy's ability to correctly detect motion.
Please let me know your feedback for the new beta.
Should we be resetting the "1s" or "2s" counters as well?
JT
@DsrtRngr - Thanks for your feedback. Are you saying that you found some real events were missed, and you had to increase the sensitivity because of this? What beta version are you using (b5, b6, or b7)?
@jtodd - Yes the new beta reset the sensitivity to 50 automatically when coming from an older version. I'd strongly recommend the 2s trigger time option to eliminate as many false-positives as possible. Please report back about how it performs, and if you can send movies of any false-positive detections that would be great.
I have a driveway and sidewalk that is 8-10 meters from the house. We have people walk our block at midnight, and I use that as my baseline. If all 4 forward facing cameras get them, I feel confident that I didn't miss any action for the rest of the night. (2 analog, 1-QCN8001, 1-QCN8009).
I have all of my cameras on 75 and even though it was windy yesterday I only got two false-positives because the large tree the camera is mounted on moved in addition to trees in the picture. Normally, with them set at 75, I would just get continuous motion capture, hundreds of files. In addition, I am capturing new shots of wildlife at night.
=
chip
SoCal
One of the improvements with the new algorithm is the fact that SecuritySpy is now measuring the amount of background noise on a per-pixel level, and using this to set a threshold (for example 4x this noise level) above which to count that pixel has having motion, when a larger change happens. However, this threshold wasn't being adjusted based on the sensitivity setting, so even at high sensitivities your motion wasn't consistently exceeding this threshold level.
So I have just posted a new beta that does adjust this threshold based on the sensitivity setting. So you will find that at the higher sensitivities, more movement will be picked up as real motion, even in poor lighting conditions. So if you are able to test this one please let us know your results.
However, this will certainly increase the chance of false-positive triggers. You will get better results with better lighting - for example installing an Infra-Red LED light panel to illuminate the scene at night (these are cheap and easy to install).
How about this for a "feature request"... on the camera settings window (ctrl-K), what if each camera setting had a radio button for "Windy" conditions. Then we could toggle back and forth between these histogram maps you're building. Likewise, there could be a separate one for day/night. ? Then, make them scriptable so day/night or windy/normal can be automated?
@DsrtRngr - the changes in b10 should help your situation, but the images you are getting from your cameras at night are simply too noisy and the people walking/jogging on the sidewalk are too far away and too dim to be captured reliably (specifically looking at your NE56 camera here). In the clip of the jogger you sent, he is virtually the same lightness as the background, and too far away to distinguish any details (clothing, facial features etc.) This isn't a specific problem with the camera itself - in fact the image is reasonably good, it's just that it's too wide angle to pick things up at such a distance with the insufficient lighting available. With a high sensitivity setting (70 or so) and a short trigger time, you should be able to capture such motion, but at the expense of a lot more false positives.
Thank you all for your help with testing and sending example movies, this has been much appreciated and has all helped to significantly improve SecuritySpy.