Safely Back to Work
Edge devices having capability of processing Deep Learning models can be used for on premise processing of the data . Edge here refers to the computations that will be performed locally on the device itself . The camera can be directly connected to the compute device which can capture and infer frames in near real-time . An ideal device for this kind of processing would have an onboard GPU for low latency inference and good memory for local computations .
LiDAR
The LiDAR is placed on the ceiling or on a wall so as to capture a wide area and angle . Similar to how LiDAR would be used to calculate occupancy , moving objects will be assigned a unique ID and with a multi-LiDAR set up , the data of the moving object will be aggregated and tracked across multiple LiDAR zones , appearing on the LiDAR software platform as one object . Each object will have normalized x , y , z coordinates across all zones to indicate a specific position at a given timestamp . The LiDAR software platform can perform simple math operations to compute distance between objects . By measuring this distance and setting thresholds so that it only records moving objects that are people , one can measure that the distance between individuals are at least 6 feet in order to ensure compliance with social distancing guidance .
IMPLEMENTATION CONSIDERATIONS
Video Analytics for Occupancy Detection
Implementation of video analytics methodology for tracking occupancy requires proper placement of the camera so that its field of view is maximised . The camera needs to be mounted at a height of about 6-7 feet above the ground so as to get a wide field of view . At the same time , it needs to be ensured that people ’ s faces are clearly visible in the feed captured by the camera so that occupancy can be tracked accurately . The camera resolution , lighting condition in the area and camera position all play a significant role in this . Moreover , if there is more than one camera installed in one closed space such as a room , the overlap in their fields of view has to be accounted for to get an accurate count of occupancy .
Video processing is a processor-heavy task , and hence a good amount of processing capability will be required to implement this method locally for real-time occupancy tracking . A workaround for this is the use of cloud services for processing , but this is bandwidth heavy and also gives rise to some concerns related to privacy of data and / or have compliance issues with privacy regulations depending on where the data is being stored .
Video Analytics for Social Distancing
As video analytics relies on camera feeds which in turn relies on other factors such as lighting condition , pixel quality of image , position of camera , overlapping objects , etc . this solution does have some limitations .
IIC Journal of Innovation - 31 -