Skip to content

Run captured videos through trigger test

edited June 2021 in SecuritySpy
As a new user, one of the things I am finding the hardest is setting the correct trigger "recipe". What I think would make this easier is if I could run various previously captured files through the trigger to see if they would cause the trigger. Basically, I would set it so it would capture a lot of files based on little motion, and then I would re-run the files through the system, tweaking the trigger recipe until it just triggers on the files I would have wanted it to. If it were really smart, the system could even tweak the settings itself. I could provide a hundred files, marking 50 that I want to pass. The system would then run them until it find settings where the 50 pass while the other 50 don't.
I assume this is kind of what you guys do with captured files provided by customers to test your code out.
I know this is reach, but I would thought I would just throw it out there as an enhancement idea.

Comments

  • Thanks for the suggestion. It is a good one in principal, but in practice it wouldn't always work so well because the motion detection algorithm gradually adjusts to the detected level of noise across the frame, and uses this noise level to decide on the threshold of change above which it marks a particular part of the frame as having movement or not. This adjustment is vital because the amount of noise in the image in low light is far higher than the amount of noise in good light, and this adjustment allows good motion detection in both circumstances. This noise detection is gradual by design, and may take 10 seconds or more to calculate.

    So, if you present a pre-recorded movie file to the algorithm where the motion event starts very near the beginning of the file, you may get a different result from when the algorithm is working from the continuous realtime stream where it has a long time before the motion event to calculate the base conditions.

    The best way to tweak the settings for good motion detection is to look at the Preferences > Cameras > Triggers window when there is some activity in the stream. Here you can see which parts of the image are causing motion detection, which will allow you to adjust the mask. Then, if you are finding that you are getting too many false-positive detections, the best thing is to increase the trigger time. On the other hand, if you are finding that you are missing real events, the best thing would be to reduce the trigger time and/or increase the sensitivity.

    These two blog posts should help:

    How To Achieve Effective Motion Detection

    Optimising SecuritySpy’s AI Object Detection
Sign In or Register to comment.