It would be very easy to use motion with the new version too, you could just install it and use the config-file that is used at the moment.jbeale wrote:Thanks for the permission change update, so at least now there is a working version with 'motion' even if it doesn't get followup support after the motion-vector release. (Right now the install script seems to auto-update to latest; is there any mechanism to request installing a previous version?)btidey wrote:I have added those group / permission changes into the repo. Thanks.
(...) I think the idea is to get it as good as current motion as quickly as possible. It would get messy to support both modes. Currently, the motion detect branch has stopped using external motion. However, we could revisit this if it looks like we can't get the equivalent.
I understand it is a major pain to support two versions and I agree it's reasonable to abandon 'motion', if the replacement is functional; even better if a user could somehow choose at the 'install' step the old/unsupported/not-updated (but still working) 'motion' version.
I've currently got three systems running RPi Cam Web Interface with motion 24/7. If you'd like beta testing for motion-vector, let me know
In the past I actually made my own flavor of motion with a very simple algorithm all in Python, with the picamera library. It mostly works but I decided I needed to do true "object detection" instead of per-pixel change detection to make it better, and that would be more complex. https://github.com/jbeale1/PiCam1
However, with the motion vectors there are way more possibilities, low cpu-ussage and 30fps-detection possible, so I think with a bit more development this solution will be much better.
About the progress: I changed the structure of raspimjpeg completely to support motion detection without buffering, at the moment the h264-encoder is fed by the preview-port of the camera instead of the video-port. There is also a splitter now right after the preview-port to support jpg- and h264-encoding at the same time. Additionaly, the structure is changing dynamically after certain events (e.g. capture started: reconnect h264-encoder from preview- to video-port). However, I'm still strugling with errors despite many many hours of debuging, so it might need a bit more time to publish the new version.
The current progress is in the MotionDetection-Branch on github as Robert mentioned, so if you want to test it, just clone the branch, remove the git pull command in the installer and install it over your current version.
It would be very useful if some of you could test the new version and give tfeedback here.