Making Optic Flow Robust to Dynamic Lighting Conditions for Real-Time Operation
Abstract
Existing optical-based state estimation algorithms fail to adequately handle dynamic lighting changes ubiquitous in outdoor environments. To overcome this shortfall, we propose a novel approach that is applicable to all optic flow algorithms, allowing them to operate in dynamic lighting conditions at operational tempos. We posit that the use of a preprocessing filter based on the double derivative of the image will substantially reduce the instability caused by dynamic lighting conditions and improve the overall accuracy of position estimates without a substantial loss of information. Our preprocessing step does not significantly add to the computational cost and requires no a priori knowledge of the environment. In this report, we compare the results of optic flow with and without use of the filter, showing that the former yields a significant improvement in position estimation accuracy as compared to optic flow calculations carried out with a standard input. Preliminary experiments demonstrate the potential of the proposed methodology.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 17, 2016
- Accession Number
- AD1005369
Entities
People
- Allison Mathis
- Daniel Donavanik
- Jared Shamwell
- Joseph Conroy
- Ryan Robinson
- William Nothwang
Organizations
- United States Army Research Laboratory