Page 1 of 1

Re: Zoneminder Darknet yolov2 tiny Hack

Posted: Fri Jan 19, 2018 3:47 pm
by Paranoid
operat0r wrote:
Fri Jan 19, 2018 2:24 pm

* I want to run the script on the path after an alarm is triggered
* I am using -120 seconds because I'm too lazy to do it right (IE just looking for modified files instead of actually knowing what JPG's to check)
* I wondered if I can some how turn on degbug logging or something so I can know the alarm frame image names to run darknet on
* I am using tiny yolo with NO GPU as I need 4gig card to save me 9 seconds and 2-3% accuracy .. SO GPU IS NOT WORTH it for just basic "person" detection. (ref google groups post )
Look at zmtrigger.pl. If you enable it then you can telnet to it and it will tell you when events start and stop and what the event number is.
With the monitor number and the event number you can work out where the images will be stored.

Here's an example output

Code: Select all

1|on|1516375720|2927992
1|off|1516375727|2927992
3|on|1516375845|2927993
3|off|1516375848|2927993
2|on|1516375849|2927995
3|on|1516375849|2927994
3|off|1516375852|2927994
2|off|1516375854|2927995
If you turn on analysis images (CREATE_ANALYSIS_IMAGES) then all you need to do is look for images called *-analyse.jpg and corresponding *-capture.jpg is the one that alarmed.

Re: Zoneminder Darknet yolov2 tiny Hack

Posted: Fri Jan 19, 2018 4:23 pm
by rockedge
here is a thread some of us were on using darknet and YOLO . I was using Darknet/YOLO to analyze individual Frames. Also messed around trying out opencv to determine the direction of travel in an motion event in ZM. I am busy at the second but will be back to share some more stuff if your interested.
viewtopic.php?f=36&t=26222