Making Automated Training Contributions

Making Automated Training Contributions

 In Brief

The camera you are adding can automatically take a snapshot of a motion event that can be used for training.

Providing contributions to the camera training fleet requires no effort and helps scOS to make your system, and other customer systems, more accurate at picking up relevant motion (with MotionX) - the entry point on detecting crime or knowing how and when to personalise your home.

Specific data will be used from the point a contribution is enabled - no past data is used.

You can select from three domains:

  • MotionX (motion detection)(in active deployment).
  • ObjecTron (recognising objects)(in active deployment).
  • Syne (recognising people)(in development).

Please read more about these options below. Data is not transferred between domains.

Contributions will be recorded for a potential competition in the future, however, this is down to our discretion on when we do so.

Terminology

  • "Submit" means that the images arrive, encrypted, into a dedicated dataset that is stored in Amazon Web Services (AWS) Ireland region. Why Ireland? This is where the most feature capable data centres are located.
  • "Core Architecture" refers to a set of AI models that work together and run in scOS Cloud.
  • "scOS Cloud" refers to processing performed in AWS Ireland or AWS London.

In Detail

    Domains

    MotionX

    Imagery will help train MotionX - our innovative motion detection system. Your Intelligence Hub will submit 10 seconds of video at 5FPS (as images) where it has low or likely confidence that there is currently motion. Classification is performed by the scOS internal team - placing frames in either noMotion or importantMotion.

    ObjecTron

    Images will only be used for annotating objects (people, cars, packages etc) that are useful for scOS Intelligence understanding the context of event. These annotations are done manually by the scOS internal team where they draw boxes over objects for their detection.

    An image has a 0.5% chance of being submitted if there is motion, and 10 minutes has passed since the last submission, from your Intelligence Hub. This could equate to 2.5 images being submitted each day, depending on activity.

    ObjecTron runs inside of the Core Architecture.

    Mnemosyne (Syne)

    In development. Images of known people will be used to build our Syne model that recognises known people just like you would and ensure unrecognised people are not mistakenly identified as known people. We don't use facial recognition due to lack of accuracy.

    Syne requires a vast and distinguishable, from person to person, set of imagery to recognise people based off from patterns like types of clothing worn, posture, skin and hair colour, hair style. These attributes are impacted by lighting and weather conditions, time of day, camera quality, image colour (night vision or not). Across a year, all of these factors change and Syne needs to understand that while these change, the person can still be recognised.

    If enabled, submissions originate from our Core Architecture in scOS Cloud, when ObjecTron determines people objects in an event. For the people that are known and recognised, Core Architecture will submit one image randomly throughout the day, every two days, if they were seen.

    Our goal here is to build a broad representation of a person for accurate recognition. Of the images submitted, we will filter out similarities, only 80 will be randomly used, based on the average person having between 120 and 150 items of clothing.

    We may decide that 80+/- are needed based on results. It could be a while until we have enough data for this feature to be trained and released within our Core Architecture.

    Where Is This Data Stored and For How Long?

    All contributions are securely stored on AWS in the Ireland region (along with our other processes). Access to this data is strictly controlled and limited to authorised personnel and follows the Principle of Least Privilege. The data is housed in encrypted storage mediums to ensure maximum security and compliance with relevant data protection regulations.

    You own your data and only allow us to permit use of it. Through the Data Integrity Zone, in CaelusView (in development), you will be able to view all of this data on scOS. There you will be able to selectively delete all or some of your contributions. Otherwise, as default to choosing to contribute, data is retained for as long as it remains relevant to improving and evolving the capabilities of the service. This period may vary as technological and operational needs develop. On a yearly rolling period, we will reach out to you to ask if you are still happy contributing.

     

    The scOS community is grateful for your contributions for building a better network that proactively combats crime!

    -

    We reserve the right to update this page in respect of the shared improvements to the network. Any significant changes, like changes in the purpose for collecting data, we will notify you, and where a change is made, it will only apply to data collected onwards from a publicised date.