Releases: dcmartin/motion-ai
Birthday Release 2020 (v3)
Birthday Release 2020 (v2)
Well, I forgot some parentheses.
Birthday Release 2020
We continue to work in both the engine room to improve performance, eliminate bugs, and connect up the new devices on the Lido deck.
On the Lido deck we have improvements in design and decor with more icons, better naming, and more exploitation of images being worth more than 1000 words.
The system now also provides top-level information on system operation, including AI's on-line, Messaging operational, etc..
Notifications now include images and those images are linked to the animated event video; I think it will work through the Nabu Casa and Web interfaces externally, but who goes anywhere these days?
Cameras may now be specified with an entity to watch, e.g. "dog" or "car" on the "interiorgate" camera and corresponding notifications can now be enabled per camera for the designated entity.
Person, Vehicle, and Animal views formerly on the Overview and using masked images has been moved to the Administrator view and unmasked. These views will be used for inspection and selection of new "ground truth" in the training loop.
Apple iOS application integration works properly, including association with Apple watch.
Simplify, simplify..
Continue to "cut the grass for the goats to eat" with improvements in the display of information.
Both visitor and administrative users now have overlays for information about the most recent detection and associated person, vehicle, or animals in those detections. These specifciations enable the counting of those entities (person, vehicle, animal) across all cameras; statistics for 1m, 5m, 15m, and 30m are also calculated.
Some additional cleaning and pipe fixing in the engine room to ensure parity of implementation across entities and across cameras.
All devices now emit their LAN IP address as part of their configuration information on MQTT; the IP addresses are collected and made options for interrogating the Motion Classic addon through its REST API; see the SYSTEM view for administrators.
The cameras are now enabled with specific entity detection parameters; all default to 'person' but any entity supported by the YOLO AI may be specified (n.b. tiny only detects 20 entities; the others detect 80); the AI options are dynamic based on the YOLO configuration.
FACE and ALPR continue to lag YOLO w.r.t. UX, but underlying plumbing is highly similar; however, additional integration is required between FACE and Person and ALPR and Vehicle detection.
FWIW: The last "working" version of HA is 0.116.4 w.r.t. the utilization of "non-conformant" REST API calls (e.g. YOLO configuration).
The system has been tested with v0.118.4 and outside of the non-conformant REST API issue, all else appears to work as designed.
Updates for SRE
Added support for devices running the motion addon to provide their IP address in the MQTT payload.
Updated UX and YAML to process.
Simplified dashboard
The "dashboard" (aka the first view for either administrative or normal accounts) now includes information that should help debug start-up as well as help w/ identifying cameras which go off-line.
The "add-on" referenced is the Motion Classic (motion-video0) add-on in github.com/dcmartin/hassio-addons repository. It has REST API end-points for both its configuration (cgi-bin/config) and status (cgi-bin/status). Those are now captured as REST sensors and used to determine if the addon is configured and running.
The MQTT broker is also being monitored for activity greater-than zero (0) during the previous 1 minute; if no traffic occurs, then MQTT is not connected.
The YOLO (and soon face and alpr) has a configuration REST API, but the current 0.117.6 server is picking up an "upgrade" from the underlying HTTP libraries to find the output invalid (n.b. closed unexpectedly); IDK why, works w/ curl, but it's on the list. Therefore the status for YOLO is currently an MQTT sensor listening for the announcement when the AI starts; however, this sensor does not indicate if it is still running (hence the REST API). However, it should help with start-up.
LMK if you have question or need assistance in installing; best to find help on FB or Slack.
Home Assistant v0.117.6
The latest version of Home Assistant (v0.117.6) is working properly for Motion-AI.
More automation
More automation for installation and configuration!
Clean-up icons, naming, ... for UX.
Cameras now automatically discovered from Motion Classic addon configuration and from MQTT traffic.
Improved installation (HA v0.116.4)
This release automates installation to include the downgrade of Home Assistant core to version 0.116.4 and configuration of default camera as locally attached on /dev/video0 using the same host for MQTT, YOLO, FACE, ALPR, and the Motion Classic addon.
Updates for Home Assistant (v0.116.4)
This update is in response to changes in v0.116, but before changes in 0.117.
This software works with v0.116, not v0.115, and not v0.117.
Recommended version of Home Assistant is v0.116.4