As we get acclimated to being homebound to combat a viral pandemic, we have begun to internalise that this may be one of several changes we have to adapt to in the coming weeks. With lockdowns enforced in many places across the globe, people have been keenly following the news cycle to understand the measures being explored to ease restrictions.
Topmost on everyone’s minds has been the imminent transition back to the workplace.
While governments in some countries have allowed the reopening of offices – albeit with some conditions – it cannot be emphasised enough that we will all be joining an altered version of our workplaces once the lockdown has been revoked.
With social distancing being the key tool and an effective measure as part of the ongoing effort to mitigate the spread of Covid-19, are workplaces ready to have the same upheld within their establishments?
AI To Check On Social Distancing
Some companies, including Amazon, have been leveraging artificial intelligence (AI) and machine learning (ML) software to monitor the distances maintained between their warehouse staff. These tools flag warnings in real-time when expected behaviours deviate from a certain standard and cause aberrations.
Drishti, a startup based out of Silicon Valley, develops AI tools that joins a growing suite of technologies that are increasingly being used to surveil workers. Another notable example is Landing AI.
Founded by data science influencer Andrew Ng, this startup has created an AI-enabled system that can be doubled down as a workplace monitoring tool to complement efforts to maintain social distancing protocols in offices. According to the California-based company’s blog, it analyses video streams from mounted cameras to detect if employees are keeping a safe distance from each other.
The Landing AI social distancing detector demo visually explains the approach employed by the startup to achieve this. While the left shows people walking down a street, the right offers a bird’s-eye view of the same, with each person represented as a dot. These dots turn red when they move close to each other, indicating when social distancing protocol has been violated.
If companies integrate this software into their security camera systems in a work setting, it could be used to highlight people in danger of possibly contracting the virus, or even spreading it further. According to the startup, it can also be used to issue an alert to remind people to keep a safe distance.
What is more, since AI is continually watching these video streams, it can extract valuable data about timing and actions, which can in turn, inform new ways of assigning work and rearranging workspaces to prevent such violations.
Technical Methodology Behind Landing AI’s tool
It comprises three main steps – calibration, detection, and measurement. According to Landing AI, the tool must first be calibrated to map any security footage against real-world dimensions. Next, a trained neural network identifies the people in the footage, while another algorithm computes the distances maintained between them.
Calibration: This includes shifting from perspective view to a bird’s-eye view. How is this achieved? When taken from a single camera, select four points in the perspective view and map them to the corners of a rectangle in the top-down view. Although this method can be challenging to apply correctly to the entire perspective image, Landing AI claims to have built a tool that can enable non-technical users to also easily calibrate.
Original perspective view overlaid with a calibration grid (left). Bird’s-eye view (right)
Detection: This is followed by applying a pedestrian detector to the perspective view to draw a bounding box around each one. According to the blog, the startup has used an open-source pedestrian detection network based on Faster R-CNN architecture.
Measurement: The last steps involve estimating their location in the bird’s-eye view using the bounding box for each pedestrian. Finally, compute the distance between people and scale the distances by the scaling factor estimated from calibration.
As critical as it is to maintain safe distances between colleagues in a work environment, it is just as important to be mindful of these practices when commuting to work.
Deep learning algorithms, computer vision and image processing can be used to measure social distancing in public spaces. For instance, Newcastle University’s Urban Observatory has developed algorithms that can automatically measure social distancing between pedestrians on the road.
Innovations like these can help identify bottlenecks where social distancing is difficult to maintain so that appropriate strategies can be implemented. What is more, it also affords insights into long-term behavioral changes.