Obstacle detection algorithms in thermal infrared images for collision avoidance

Obstacle detection is an essential component in any Sense and Avoid system. Thermal Infrared (TIR) imaging technology can provide enhanced capabilities to current vision-based SAA systems, allowing for operation under extreme illumination conditions, such as direct sun exposure, and during the night. This paper presents a comparison of image processing algorithms for the detection of flying obstacles in TIR images, including both methods based in morphological operations and methods based in the Discrete Haar Wavelet Transform (DHWT). To the author's knowledge, the use of these methods for obstacle detection in TIR images is completely novel. These algorithms have been evaluated on a dataset of more than 5K image pairs, including both ground-based and airborne TIR images. This dataset includes images captured under very different illumination conditions (normal exposure, direct sun exposure and at night) and with various sky conditions (clear and cluttered skies). Conclusions validate these algorithms by presenting measurements in terms of precision and recall for the detection of flying obstacles in several video sequences and a study of the computational cost of the evaluated algorithms.