- We’re rolling out updates to fine tune sensor accuracy. You may notice up to a ~5% reduction in utilization metrics from 3 January 2023, as a result of the changes.
- You can now view the cost of unutilized work points in your currency of choice (USD, GBP, EUR, AUD, and NZD) on your dashboard.
In today’s feature spotlight, we reveal a new update to sensor accuracy and break down how sensor accuracy is measured and optimized through machine learning at XY Sense.
We also introduce new localized cost metrics on dashboards which allow multinational teams to easily track and compare the costs associated with underutilized spaces in local currencies. View this change live in your dashboard.
Let’s jump in!
Our ongoing commitment to accuracy
We’re incredibly proud of the accuracy of our market-leading Area Sensor which captures anonymous sightings down to 1 ft (30cm) every 2 seconds to deliver workplace intelligence way beyond simple people counting.
However, no privacy-preserving sensor is ever 100% accurate, 100% of the time. That’s why, alongside the support we provide to customers during installation to facilitate optimal sensor positioning and data capture, we continually invest in research and development of our proprietary machine learning algorithms to deliver ongoing accuracy improvements and nudge towards the nirvana of 100% sensor accuracy.
How do we measure and improve sensor accuracy?
We measure accuracy by looking at the number of false positives (incorrect detection of a person), false negatives (person not detected), and positional accuracy (accuracy of the detection location). This is done through experimentation in the XY Sense labs and in select test office environments by comparing XY Sense occupancy readings to footage collected through separate, independent video recording systems. Any inconsistencies help us to identify opportunities for further training and augmentation of our machine learning algorithms.
Once developed, new versions of our proprietary machine learning algorithm are then rigorously tested in both our labs and with clients involved in our beta programs to prove demonstrable improvements to data accuracy before software updates are remotely released to clients across our global network of sensors.
We’re excited to announce that we’re now ready to release a major algorithm update we’ve been working on which further improves sensor accuracy.
Area Sensor Accuracy Upgrade
On January 3, 2023, the latest version of XY Sense’s motion detection algorithm will be released. This firmware update will improve the accuracy of XY Sense data collection delivering more accurate reporting for workplace teams in 2023.
- No action is required from customers to support the sensor software update.
- The update will occur remotely outside of business hours with no anticipated sensor down time for this update. As such, there will be no gaps in data collection.
Key features include:
Filtering sightings using motion algorithm
Our new motion detection feature reduces inanimate object detections that are incorrectly detected as people (false positives). The upgraded algorithm invalidates sightings when no movement is detected across a series of sightings. This results in more accurate occupancy reporting, analytics, live views and real-time integrations.
Impact on analytics
Once sensors are updated you may notice a slight (~5%) reduction in utilization metrics across dashboards as occupancy detections are fine tuned in each workplace location.
For this reason, we have a planned rollout on 3 January so that your 2023 dataset starts the year on the new update.
For API Integrations, the new updates will reduce overall event volumes in our webhooks and reduce the volatility of sightings in the Sightings API. Lab tests indicated a 40% reduction in event volumes, reducing the noise for our API integrators.
In R&D: Reduced duplicate detections
Our team is working to reduce duplicate occupancy detections with a new proprietary single-shot inference model. Instead of processing the raw sensor sighting in 9 sections where an occupant can be detected in multiple sections, our new model helps the sensor process a complete view of the scene, eliminating this challenge. Early test results are exciting – we’re seeing a 22% improvement in true positives, a 45 % improvement in false positives, and a 90 % reduction in duplicate sightings in our test dataset (which contains more than 100,000 scenes).
In R&D: Additional classifications in our model
We’re also training our sensors to recognize more everyday objects around the workplace. This feature reduces false positive readings by enabling our Area Sensor to distinguish objects such as plants and chairs in addition to humans.
Localize the currency of your cost metrics
We’re proud to be helping teams in over 15 countries take a data-driven approach to workplace strategy.
New localized cost metrics on dashboards now allow teams to easily track and compare the costs associated with underutilized spaces in local currencies, including; GBP, EUR, AUD, NZD and USD (default).
This feature is now available on the homepage and in the Portfolio dashboard.
Log in to see the new currency localization update or contact support for currency requests or changes.