- Landing AI built an AI Social Distancing Detector to measure space between people and integrate with CCTV to enforce workplace distancing.
- Demo video shows red boxes for people too close and green boxes for those in a safe zone.
- Calibration converts camera images to a bird's eye view using four point mapping and assumes everyone stands on flat ground.
- Detection uses open source detector and computes scaled distances; company says it does not identify individuals, but privacy concerns were raised.
Over the past few months, social distancing has proven to be a very effective way to curb the spread of coronavirus. Although a lot of people are under lockdown, many people, including healthcare and pharmaceutical industry workers, are still out there working to save lives. As such, Landing AI has come up with an AI Social Distancing Detector tool to help them ensure social distancing at their workplace. Also, the idea is to integrate the AI Social Distancing Detector tool into closed-circuit TV systems, measuring the spaces between two people.
How The AI Social Distancing Detector Works
Two weeks ago, Landing AI posted a demo video on YouTube, analyzing how the AI social distancing detector tool works. The video was performed to give a visual explanation of the approaches which the AI social distancing detector will utilize.
https://www.youtube.com/watch?v=15iIV1Lff-M&feature=youtu.be&app=desktop
According to the video demo, the red box highlights people who are not keeping a safe distance. And the green box shows people who are in a safe zone.
Let’s have a quick look at the steps involved in using the AI social distancing detector below:
- Calibration
The first step, which is calibration, involves converting the homography that alters the images into a bird’s eye view. No doubt, the input frames are taken from a single camera. Because of that, the system assumes that everyone is standing on the same flat ground. It then calibrates by choosing four points in the perspective views and mapping them to the corner of the bird’s eye view.
Furthermore, there’s an inclusion of a lightweight tool by the researcher, to assist non-technical users with real-time mapping tasks.
- Detection
The second step of the AI social distancing detector pipeline involves applying an opensource pedestrian detector to the perspective views. That’ll help to draw a bounding box around each pedestrian. However, getting the best fits involves applying minimal post-processing, including setting real-life rules.
- Measurement
The last step of the AI social distancing detector involves two approaches. The first approach involves taking the location estimates of each person in the bird’s eye view. The final part involves computing and scaling the distance between every two people from the bird’s eye view.
Although the company has emphasized that the AI social distancing detector tool doesn’t recognize individuals, many people have expressed their feelings, by raising privacy issues under the post on YouTube.
