Abstract We present a new end‐to‐end pipeline for the detection of Martian dust devils in rover imagery that substantially expands current capabilities for atmospheric monitoring. Applied to 23,409 images from the Perseverance rover across 200 sols, our system identified 19 dust devils, 18 of which were previously undocumented. This represents a significant increase over prior manual detections. These results enable systematic searching of data covering a longer time span and provide quantitative characterization of dust devil morphology through stereo photogrammetry, including distance, height, and width measurements. The pipeline combines a custom object detection framework with novel preprocessing components, including a Random Forest based prefilter and a frame differencing module. These elements enhance sensitivity to low‐opacity, amorphous vortices that are challenging for traditional detection methods. By focusing on faint motion features across time, our method detects events that are often overlooked by both human observers and standard tools. This work introduces a scalable framework for real time detection of atmospheric activity on Mars. The findings support improved scientific understanding of dust devil behavior and also offer practical guidance for future rover design. We advocate for greater integration of environmental sensing, stereo imaging, and onboard machine learning in future missions.
Hatfield et al. (Fri,) studied this question.