MV Intelligence Training
Overview
All second-generation (MVx2) cameras are capable of processing powerful analytics on the camera itself and transmitting this metadata to the Meraki cloud. This revolutionary architecture dramatically reduces the cost and complexity of gathering detailed analytics in any environment.
One of the new and exciting capabilities of this platform is the ability to do machine learning-based analytics. With this comes object detection of persons and vehicles. This allows you to understand the occupancy and use of your environments and search for specific security- or emergency-related events. In order to make sure this object detection is robust and trustworthy across a wide variety of locations and scenes, we have built MV Intelligence Training.
What is Intelligence Training?
MV Analytics serves many advanced use cases today, built on top of detecting people and vehicles. In order to improve the current capabilities of our machine learning models and introduce new capabilities in the future, MV needs to train on visual data from a wide variety of scenes.
Intelligence Training enables customers to register their MV cameras to contribute to and benefit from this model enhancement lifecycle, while providing them tools for transparency and control.
Always striving to improve
The goal of MV analytics is simple: bring as much intelligence and capability to the edge of the device and reduce the need for as much infrastructure and complexity as possible. By enhancing our capabilities and performance at the edge, insights from the camera can be used to deliver real-time results for security and business intelligence needs alike.
The accuracy of the object detection is dependent, in part, on the ML model’s understanding of the environment.
Diversity of data is key
To improve the quality of our machine learning analytics, we need to provide training samples from a diverse range of environments; indoor and outdoor, brightly lit or candle lit, rain, sleet, and snow. We also need to provide a diverse set of images for the objects we're interesting in monitoring or tracking across these range of environments. People of all shapes and sizes, outfits, accessories, and poses. Vehicles of every type, size, color, etc. By expanding our machine learning analytics' understanding of the breadth of what is a person or vehicle, we enhance it's ability to perform this task reliably for *every* setting.
Privacy by Design, Privacy by Default
The privacy of our customers' data has always been at the forefront of our product design principles. We have placed extensive engineering and development of our entire camera platform to ensure things like video encryption end-to-end, signed firmware, secure boot with trust anchor module, two-factor authentication, and many other security- and privacy-centric features are the standard, not the exception. Our approach to customer video is no different.
- Privacy first: MV Intelligence Training is off by default and is only configurable by Organization Administrators. We put customers in the drivers seat for what information they choose to share and from which devices.
- Ease of use: The process to opt cameras into Intelligence Training is simple and provides complete control over the data they share.
- Bandwidth conscious: We only collect what we need, and nothing more. Changes in motion, lux, or object detections flag interesting frames for us to analyze.
How to configure Intelligence Training
There are two methods initially available for configuring cameras to opt-in to the Intelligence Training process: the temporary opt-in banner and the Intelligence Training page.
Temporary opt-in banner
In order to provide visibility and additional detail on this new feature, an opt-in banner has been added to all Organizations. This banner is only visible to Organization Administrators and provides context for the purpose and value this new feature provides to Cisco Meraki MV customers. This banner will disappear after it has been dismissed three times, or immediately for all Org Admins of the network if a camera is registered into the service.
Intelligence Training page
Located under Cameras > Configure > Intelligence training, the Intelligence Training page gives users complete visibility and control over which cameras are opted into the process and the methods for removing data collected or opting out.
Opt-in cameras to the service
There are two flows to arrive at the opt-in tool: either via the temporary banner (described above) or the Intelligence Training page. Regardless of which flow, users can opt-in cameras by following these quick steps:
- Select the camera(s) from the thumbnail or list view
- If you have many cameras, you can search by camera name, model, or serial number
- Click the "Confirm Selection" button at the bottom
- Review the Automated Collection Agreement (downloadable and accessible always from the Intelligence Training page)
- Confirm by hitting the "Agree and Opt-in" button that appears below
Remove collected data from a time range
In the event training data may have been shared from a time range that shouldn't have (for any reason), customers have the ability to remove any data from that range by following these quick steps:
- Select the camera(s) from the list
- Click the "Remove Frames" button that appears
- Select the appropriate time range where data should be removed (or select the checkbox to remove *all* collected data)
- Confirm by hitting the "Remove Frames" button that appears below
Opt-out cameras from the service
You can remove registered cameras from the service by following these quick steps:
- Select the camera(s) from the list
- Click the "Opt-out Camera(s)" button that appears
- Confirm by hitting the "Opt-out Camera(s)" button that appears below
Monitoring and logging
By navigating to Camera > Monitor > Video access, Organization Administrators can observe all automatic collections taken from cameras for Intelligence Training, along with start/stop timestamps. You can filter for these events directly by selecting the type filter as "Automated training export created."
Technical FAQs
How does it work?
Every Meraki camera sends metadata for things like object detections, motion, and light levels. Once a camera is registered into Intelligence Training, software on the Meraki Backend monitors for interesting events from cameras using only this metadata. This software will selectively collect a few frames for each interesting event and send this to our secure cloud storage for automatic annotation and training. Data is encrypted at all stages of the process and strict access policies are in place across all resources. For information on our datacenters, storage architecture, regulatory compliance, and otherwise, please see our Trust page at https://meraki.cisco.com/trust.
Can I stop sharing data from a camera?
You can choose to stop sharing from a camera or delete previously shared data at any time.
What do you mean by "machine learning"?
Meraki smart cameras use deep learning, a type of machine learning at the forefront of artificial intelligence research, to drive our computer vision object detection. The smart camera development teams continually show a computer thousands of examples of what objects look like and it "learns" how to identify them more and more accurately over time. The model improves as we provide it with additional training data.
The smart camera analytics models are trained on data that is legally owned or licensed by Cisco Meraki, and only uses data from customers who have explicitly opted-in to continually improve our models.
Does this take any extra bandwidth?
Data collection for MV Intelligence Training will consume additional bandwidth outside of normal camera operation. This process will only collect maximally ~20MB of data per hour and 200MB of data per day. We will no longer collect data from cameras after collecting a diverse set of images.