A preliminary assessment of traditional computer vision algorithms on orbital infrared frames of the martian surface

Vision-based navigation is a groundbreaking technology that is achieving unprecedented performance for various space applications. Amongst the most interesting are autonomous guidance navigation and control and hazard detection and avoidance for planetary landing, as proven by the impressive recent successes of missions such as SLIM (Smart Lander for Investigating Moon) and Mars 2020. However, the technology does have an important shortcoming: all its applications exploit visual sensors which require good illumination to work properly. Infrared sensing could help overcome this limit, but such a concept has rarely been investigated. This paper is a preliminary assessment of computer vision algorithms on infrared frames of planetary surfaces. A comprehensive set of numerical simulations is carried out to evaluate the performance of these algorithms, proving their effectiveness in providing high quality information regardless of illumination conditions. Infrared planetary surface frames were synthesized by manipulating the infrared mosaics of the surface of Mars generated with the data collected by Mars Odyssey. Experiments were conducted under varying terrain types and simulated mission-like conditions. Additionally, the effectiveness of performance enhancing strategies is investigated and overall efficiency of the algorithm candidates compared. Finally, execution times were profiled on a BeagleBone Black board to approximate embedded deployment. The study provides valuable benchmarking data and foundational insights for future research on the use of infrared imagery for autonomous navigation in space applications.