EcoSta 2019: Start Registration
View Submission - EcoSta2019
Title: A joint HMM for RFID-based object tracking in a complex environment Authors:  Kerby Shedden - Statistics (United States) [presenting]
Yang Chen - University of Michigan (United States)
Abstract: RFID technology is an inexpensive way to track the movement of objects through an environment. We present a study in which patients and health care providers in an ophthalmology clinic were issued RFID tags, allowing their motions to be tracked at high temporal resolution. The main goals were to assess the amount of time each patient spends with providers, and to quantify time spent waiting at each stage of a clinic visit. Since RFID signals are weak, and the clinic environment contains many signal-distorting physical obstacles, the raw RFID data has many gaps, and frequently places patients and providers in impossible locations or on very unlikely trajectories of motion. We developed a joint HMM for patients and providers that produces realistic reconstructions of their joint motion. The model considers both within-target behavior (e.g. the rate of transition between rooms, statistical tendencies of different patients and providers to follow certain trajectories), and between-target behavior (e.g. the number and type of targets that can simultaneously occupy a location). New algorithms were developed to jointly fit a collection of interacting HMMs, incorporating the unique properties of the RFID signal distributions and domain-based probabilistic structure. We demonstrate that this approach produces accurate and informative summaries of patient and provider trajectories and removes most of the artifacts that are present in the raw data.