Approach

Fireball detection software is one of the modules forming an automated GFO data processing/reduction pipeline to process the fireball orbital dataset captured by the camera systems. The software was capable to work in the constrained environment. An on-board Linux based computer processes the images during the day to detect fireballs by implementing the software that utilises both conventional image processing and machine learning approaches. Approximately 60GB of pictures are usually taken throughout the night before they are split into multiple tiles and processed through fireball detection modules. Tiles which only contain fireballs are sent over the mobile network to the local server. The project team studied the standardising versions procedure for software developing and testing and integrated this within the current processing pipeline. The team also integrated the data filtration tool and human review tool to clean up the fireball imagery data before training the machine learning model.