ton of video material to analyse
0ax last edited by
It's very interesting solution for my video problems, but as my starting project I would like to solve my big problem from years.
Each day camera generates abt 30gb footage
It generates one file per hour, or more, if there is a problem with network. (rather, frequently one file).
I would like to analyse entire day and put only this parts, where is motion detected by specific parameters.
After analysis, I need to throw away full-day footage as this is (now) about 5TB historical data.
So, is that possible with your solutioon CE ?
If you are recording H.264 over RTSP then i can suggest taking a look at this article http://shinobi.video/articles/2017-07-29-how-i-optimized-my-rtsp-camera
The gist of what it will tell you to do is tweak settings in your camera's internal settings, like from the camera's web panel (or with some ONVIF management tool if it is ONVIF).
This is the section of the article you should give attention to.
* Framerate (FPS) : High : 10 - 15 FPS, Low : 2-5 FPS * I-frame interval : 80 * Bit Rate Type : CBR (Constant Bit Rate) * Bit Rate : between 256kbps - 500kbps
If none of this information was helpful then you can try and save to WebM. This will use a lot more CPU per camera but it will save large amount of disk space.
Motion detection only works on live streams. Currently there is no way to audit a video for motion or objects afterward. I will consider this for future versions as it does seem to be a highly requested feature.
0ax last edited by 0ax
Thank you for your reply.
Thank you for your info about videos - but In the first steps I would like to deal with my older records.
So, I am thinking about that usage now:
Virtual Machine with better parameters - but only for this video scans, nothing more, so we can do almost everything.
- Make video stream in ffmpeg (if that is possible, but I have had seen any documentaion about it, I suppose)
- Setting it as a monitor input in Shinobi
- Is that possible ? Setting the monitor like Zoneminder "Modect" mode - where the recordings and other processes starts only when the motion is identified.
- The first problems which I see - time and timing - I have to scan about 300 days for the begginning.
a. Would be great if we can find any method to faster that, than 1:1 seconds analysis.
b. I would expect, to get something (event) and mark in the database - I would expect I can get it with exact time stamp on the day. Its the great feture which I would use, but my major problem probably :(?
Assuming that we will playing that 2x faster, that looks to be a problematic?
c. I have thought before - about spoffing real time for the virtual machine.
For that , to get time 1:1 to the day - for example:
[ would like to start scanning video material from 11-01-2017 00:00:00 ]
setting the day 10-01-2017 on vm
starting the stream
starting the whole process
and the time will be 1:1 compliant to entire day, and after 24hrs there will be next day, and I wouldn't touch it, so registering events as the January in virtual environment, as long as I have videos from January.
Is that makes sense?
And here we starting to have problems with making the analysis 2x times faster...
sorry for the late repsonse.
currently Shinobi does not have this ability.. as the event (trigger) would be recorded for the actual time. Not the time of the video.
although if it did have this ability reading at 2x would not be an issue. You can still mark the event for when it happened in the video. Just add the elapsed time to the start time of the video and bam you have an event at the correct time.
If i were to estimate the time it would take to add such a feature it may be around 4 - 6 weeks... but currently i don't know when i'll be available to even start researching ways to achieve this. If you would like to try adding the feature yourself then you can post here with questions