New AI object detection pipeline

Completely removed Tensorflow since it does not play nice with arm64 based operating systems.
Replaced with a YoloV5 implementation running via Python

Fixed MQTT error

• Better null checking in the MQTT logic

Fixed startup crash

• Downgraded get-ports

Fixed crash when not using MQTT

• Fixed an issue where the lack of MQTT config would crash the plugin.
• Updated dependencies
• Updated readme (no official support for Hoobs)

Added snapshot to mqtt

MQTT motion messages now contain a now res snapshot.

MQTT support, correctly working tensorflow binaries (C++ bindings) and many many improvements

This release adds support for MQTT messages when motion is detected.

Unifi OS fixes & non global install fixes

• Fixed Unifi OS support
• Fixed install issues on HOOBS (and other non global/root installs)

Plugin refactor

• Camera logic has been refactored, camera's are now imported into Homekit automatically, no need to add them in manually. Old cameras will need to be deleted, consult the readme!
• Support for enabling and disabling motion on a per camera basis
• Support for triggering motion
• Support for letting specific cameras act a doorbell (no two way audio)
• Support for excluding certain cameras from being enumerated into Homekit
• Preview image in the notification will now use the actual snapshot from the camera that was used internally to do the motion detection!
• Various improvements concerning Tensorflow and FFMpeg
• A lot of refactored code & improvements

Various fixes

• Rolled back TFJS-node version upgrade (I will keep working on this to hopefully eventually perform the update)
• Fixes various edge case errors
• Improved logging
• The wiki has extra information if your install fails

https://www.npmjs.com/package/homebridge-unifi-protect-camera-motion/v/0.3.3

Updated dependencies

Updated homebridge-camera-ffmpeg dependency & tfjs-node to their latest versions, and added a new postinstall script to download a compatible precompiled arm compatible (as that is not available and was the reason the tfjs-node lib was held back)