A Pattern Recognition System for the Automated Tracking and Classification of Meteors Using Digital Image Data
In this project, we aim at developing an efficient and reliable method to search through vast amounts of digital image data of the sky (at video rates of 30 fps) to identify meteors and extract associated features, such as speed, brightness (light curve) and direction. We describe a pattern recognition system that accomplishes this goal in two steps. The first step keeps track of luminous objects that follow a straight path through a set of frames; multiple features become then available for each moving luminous object. During the second step, a classification model automatically labels the moving object as either a meteor or not (e.g., or satellite); the classification model is obtained by using a machine learning algorithm previously trained on historical data. The benefit of such pattern recognition tool is twofold: 1) it obviates the process of searching through image data by eye, which is infeasible for producing a database with reliable statistics; 2) derived information can provide a better understanding of the mass distribution of meteors, which is crucial for determining the total mass flux incident into the upper atmosphere; meteors are the main source of metal ions in the mesosphere (80--100 km altitude), and can significantly affect the atmospheric chemistry of that region.
Return to oral presentation list