Physical Distancing and Crowd Density Monitoring
�
�
�
Real-time alerts for urgent situations : The real-time component supports the ability to generate immediate alerts by sending out e-mail and text messages for scenarios such as violation of the threshold for maximum number of people in a given space or not maintaining physical distancing for a long duration . Authorities can then take immediate steps to mitigate these urgent situations . Detect hotspot “ patterns ”: Hotspots are the locations and time periods where people might find it difficult to practice physical distancing . The post-facto dashboard can help to quickly identify and stop the hotspot patterns with respect to time of the day , day of the week and by location . Identifying these hotspots can help authorities take appropriate steps to prevent them in future . Cleaning Crew Guidance : Identifying areas of maximum activity helps cleaning crews prioritize the area to focus while cleaning at the end of the day . Heatmap analysis can help to quickly classify the location by activity for the cleaning crew .
METHODOLOGY
In this work , we have designed a real-time physical distance monitoring system using computer vision and streaming analytics . Figure 1 provides a high-level overview of the system . At its core , the system detects people in each video frame and tracks their trajectory across frames while anonymizing their identity . Distances between people are then calculated to determine if they are maintaining a safe physical distance . The system sends real-time notifications to alert for violation of physical distancing and generates interactive visualizations to better monitor physical distancing . For real-time execution , the main components of the system are deployed on an edge server which is running the SAS ® Event Stream Processing ® ( ESP ) engine . This solution was developed using a combination of open-source software 2 and propriety SAS software for a surveillance camera located in a cafeteria on SAS campus , referred to as SAS café from here on .
Person Detection and Trajectory Tracking
Object detection is one of the fundamental research challenges in the computer vision community . This problem is observing continuous interest from the deep learning researchers to build accurate and faster models . A major breakthrough was observed with the You Only Look Once ( YOLO ) object detection algorithm 3 which achieved state-of-art accuracy while significantly improving inference speed , thereby allowing for real-time deployment .
2
Bradski G . The OpenCV Library . Dr Dobb ’ s J Softw Tools . 2000 .
3 Redmon J , Divvala S , Girshick R , Farhadi A . You Only Look Once : Unified , Real-Time Object Detection . Proc IEEE Conf Comput Vis Pattern Recognit . 2016:779-788 . doi : 10.1021 / je00029a022 .
IIC Journal of Innovation - 3 -