Social Distancing Early Forecasting System
In this post, I will present our latest computer vision project, aiming to save lives amid the Coronavirus Pandemic. [Code]
To combat the COVID-19 virus, the Centers for Disease Control and Prevention (CDC) has recommended everyone to practice “social distancing” in public spaces, which means people should keep a distance of at least 6 feet from each other. Authorities from around the world try to enforce such guidance physically while risking infecting the officers/employees. Essential workers like bus drivers and supermarket workers have died. People also make bulky social distancing devices to battle the disease [1, 2, 3, 4, 5, 6, 7].
We’ve built an early warning system that gives warnings before people actually get too close. The idea is to identify when people are close together and then provide warnings. The system can also predict that separation limits will be breached 5 seconds before it happens, given recently observed behavior.
Our system takes the input of scene semantic segmentation features and hence helps preserve privacy while still being able to prevent “social distancing violations”. The system is based on models from [8, 9] and the processing speed can be faster than real-time speed.
Here is the visualization of system outputs. Our system yields an early alarm when a potential violation of social distancing is predicted ~5 seconds in the future. Note that our system does not need RGB frames as inputs. The RGB frames are shown only for better visualization. The videos are from public datasets [10, 11].
Our system can be placed in high-risk public locations like buses and supermarkets to remind people to keep social distancing and save lives.
Our system can be applied to different kinds of cameras.
References:
[1] https://www.youtube.com/watch?v=jNBPmcCD3b8 UK police face allegations of being ‘over-zealous’ at coronavirus social distancing | ITV News
[2] https://www.youtube.com/watch?v=VdDGUvLoJvI Coronavirus outbreak: Indian police punish lockdown offenders with violence, push-ups
[3] https://www.youtube.com/watch?v=Tsxo2hzzGfI
[4] https://www.youtube.com/watch?v=TJ_Qchx7FYU Michigan woman makes social distancing device for husband
[5] https://www.youtube.com/watch?v=p8hSSSFKv0k DDOT bus driver dies from coronavirus
[6] https://www.youtube.com/watch?v=Ot4WOR38hKs Walmart faces wrongful death lawsuit over employee’s COVID death
[7] https://www.youtube.com/watch?v=GsUjKizyiW8 Grocery Store Safety Tips During Coronavirus Pandemic
[8] Liang et al. “Peeking into the future: Predicting future person activities and locations in videos.” In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5725–5734. 2019.
[9] Liang et al. “The Garden of Forking Paths: Towards Multi-Future Trajectory Prediction.” In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2020.
[11] http://mevadata.org/
