Impact of Georegistration Accuracy on Wide Area Motion Imagery Object Detection and Tracking

Advances in sensor technologies and embedded low-power processing provide new opportunities for using Wide Area Motion Imagery (WAMI) across a spectrum of mapping and monitoring applications covering large geospatial areas for extended time periods. While significant developments have been made in video analytics for ground or low-altitude aerial videos, methods for WAMI have been limited due to lack of benchmarking datasets, data format complexities, lack of labeled training videos, and high data processing requirements. This paper aims to help advance the broader use of WAMI by evaluating the georegistration accuracy and its impact on downstream video analytics using two benchmark datasets (CLIF 2007, ABQ 2013). In addition to the current intensified interest in using deep learning for aerial object recognition and tracking, this paper motivates the need for further development of more robust and fast georegistration algorithms for multi-camera WAMI systems.

I. Introduction

There has been an exponential increase in aerial motion imagery due to advances in airborne sensor technologies, the increased adoption of manned and unmanned aerial vehicles (UAVs), and the emergence of new applications including aerial delivery, environmental monitoring, smart cities, search and rescue, disaster relief, and precision agriculture. Society is seeing a growing need for robust aerial imagery and video analytics capabilities to take full advantage of data fusion and to meet such application needs [1]. Novel methods, particularly those using artificial intelligence/machine learning (AI/ML), coupled with rapid advances in computational hardware (more powerful, lighter weight, lower energy, lower computing cost) are revolutionizing image processing, pattern recognition, and information fusion (e.g,, WAMI fusion applications [2]).