System and Method Social Distancing Recognition in CCTV Surveillance Imagery
As part of our supercluster initiative at Xtract One (formerly PatriotOne), I developed a patented system to estimate distances between people in real-time and determine their social distancing compliance. (LINK TO PATENT)
My end-to-end contributions included:
Planning, filming, and augmenting all of the in-house social distancing videos, as there were no readily available datasets at the time.
Training the PyTorch Person Detection deep learning algorithm.
Engineering the features from the detected person bounding boxes to feed into our social distancing regression model.
Training the Random Forest regression model using Python’s Sci-kit Learn library.
Packaging the regressor as an ONNX model to run on the Nvidia Triton Inference server.