This repo hosts all the code and data needed to replicate the AI detection system that was built for my MSc project: "AI for space debris detection & orbit reconstruction".
The data can be found in the following Google Drive link: https://drive.google.com/drive/folders/1fwEwQbzbYTEcPcD9p_AiPfQqSwP30uyv?usp=sharing
The complete Thesis report can be found at: https://drive.google.com/file/d/1lmV6VItuXquvpM_5oipap9nvMzqXNt8k/view?usp=sharing
The following gif visualizes the AI detection stage in a few steps:
- Records 8 frames of captured video.
- Pre-processes frames using backround subtraction and image blending.
- Runs through Faster R-CNN object detector.
- Collects predictions and plots bounding box (slighlty offset from actual, for visualization purposes).
- Centers camera payload to debris origin (assuming an appropriate gimbal system).
- Collects telemetry data, camera parameters and prediction results to calculate & display parameters of interest (e.g. debris position, speed etc.).
- Repeats for the next 8 frames.
*note the actual speed of the process has been slowed down to around 1/3 through .gif conversion