IIC Journal of Innovation 15th Edition | Page 17

Physical Distancing and Crowd Density Monitoring
Fig . 6 : This second set of heatmaps summarizes the locations of physical distancing failure , also aggregated over 5-minute intervals . This helps answer which areas have people involved in a physical distancing violation . On the left , we aggregate over 11:30am to 11:35am whereas on the right , we aggregate over 12:35pm to 12:40pm . We observe different spatial patterns between the two-time intervals , such as the increased volume of violations in the area between the two cashiers . We also observe similarities such as the consistent violations near the primary cashier station ( located at the top-left ). In practice , we build heatmaps over all time-intervals of interest and animate to show their continuity .
In contrast , our heatmap on Figure 6 is specifically designed to answer the question “ What are the locations of physical distancing failures ?” ( i . e . Which areas have people less than 6 feet apart ?) This is computed by restricting attention to only the positions of people that are involved in a physical distancing failure during the failure . This question is framed to understand whether there are features in the environment that are leading to physical distancing failures .
This analysis is performed with the highest level of privacy since the focus is not on any given individual but rather on the physical environment instead . This summary can be used to redesign the space , clear blockages , or remove obstacles to make it easier for everyone to keep their distance . In the second figure , we can see hot spots were physical distancing is frequently failing such as near the checkout areas or the area in the bottom-left corresponding to a food station . One advantage of studying these heat maps over time is that we can examine both where physical distancing failures occur and when . Our solution generates a heat map for each time interval and stitches them together into an animation video for qualitative analysis and human review .
- 12 - November 2020