Heteroskedastic Geospatial Tracking with Distributed Camera Networks

TL;DR

This paper introduces a novel framework for geospatial object tracking using distributed camera networks, emphasizing uncertainty estimation and communication constraints, and provides a new dataset for this task.

Contribution

It proposes a new modeling framework and a dataset for geospatial tracking with uncertainty estimation in distributed camera networks.

Findings

The framework effectively estimates object locations with uncertainty.

Fine-tuning improves tracking accuracy and uncertainty calibration.

The dataset enables benchmarking of geospatial tracking methods.

Abstract

Visual object tracking has seen significant progress in recent years. However, the vast majority of this work focuses on tracking objects within the image plane of a single camera and ignores the uncertainty associated with predicted object locations. In this work, we focus on the geospatial object tracking problem using data from a distributed camera network. The goal is to predict an object's track in geospatial coordinates along with uncertainty over the object's location while respecting communication constraints that prohibit centralizing raw image data. We present a novel single-object geospatial tracking data set that includes high-accuracy ground truth object locations and video data from a network of four cameras. We present a modeling framework for addressing this task including a novel backbone model and explore how uncertainty calibration and fine-tuning through a differentiable tracker affect performance.