Spatio-Temporal 3D Object Tracking in Multi-Camera Environments

In-Store Person Tracking

  • During my time at Standard AI, I worked on a novel algorithm to solve the problem of space and time association simultaneously, for the purpose of Multi-Camera Multi-Person Tracking in stores of major retailers across North America, Japan and Italy.
  • Some of the topics implemented to build this pipeline include 3D Reconstruction, Linear Program Optimization, Camera Clustering and Calibration, Hypergraph-based Tracking.
  • In comparison to the previous tracking module, I achieved a 40% increase in accuracy, and a 15% increase in speed along with adding additional scalability, which allowed the pipeline to be deployed in more than 100 stores of varying sizes.