This machine learning models uses computer vision to detect aggregate level of social distancing in waiting rooms within hospitals. The model is currently in the pilot phase at St. Paul's Hospital in Vancouver, Canada in order to ensure physical distancing and compliance to infection control procedures for the safety of our patients, staff and visitors. The model utilizes Yolo-v3 objection detection model and weights by applying a single neural network to the full image. No personally identifiable information is stored as part of the analysis. This model is not intended for medical use.

Using YOLO v3, the model uses a variant of Darknet, which originally has 53 layer network trained on Imagenet. For the task of detection, 53 more layers are stacked onto it, giving us a 106 layer fully convolutional underlying architecture for YOLO v3. YOLO v3 is a fully convolutional network and its eventual output is generated by applying a 1 x 1 kernel on a feature map. In YOLO v3, the detection is done by applying 1 x 1 detection kernels on feature maps of three different sizes at three different places in the network. YOLO v3 makes prediction at three scales, 32, 16 and 8 respectively.

Utilizing Yolo-v3 object detection model, we apply a single neural network to the full image. This network divides the image into regions and predicts bounding boxes and probabilities for human class for each region. The distance between centroid of two detected human bounding boxes are measured, with the assumption that 150cm is X amount of pixels apart in the video. The number of social distancing violation and the total number of people is recorded to a CSV file for statistical analysis as aggregate statistics. No personally identifiable information (PII) is stored.