A lightweight pipeline that detects and traces moving objects in FITS astronomical time-series using ZScale normalization, frame differencing, greedy track linking, and motion-based scoring.
Five stages transform raw FITS frames into a confirmed moving-object trajectory.
Key metrics measured on ~4000 FITS frames from the WISE survey dataset.
The high-fidelity research plot generated by the asteroid detection pipeline, showing the traced path on the FITS background.
Adjust parameters to see how the scoring function evaluates a candidate track in real time.
How threshold τ and MAX_LINK_DIST affect track count and quality (Table III from the paper).
| τ (Threshold) | MAX_LINK_DIST | #Tracks | Best Track Length | Observation |
|---|---|---|---|---|
| 15 | 10 px | ~210 | 12 frames | More blobs; higher noise risk |
| 20 ★ | 15 px ★ | ~148 | 18 frames | Default — balanced performance |
| 30 | 20 px | ~87 | 14 frames | Fewer blobs; may miss dim objects |
| 20 | 5 px | ~320 | 8 frames | Under-linking; many short fragments |
| 20 | 30 px | ~104 | 21 frames | Over-linking; wrong associations possible |
Detecting and tracing small moving objects in astronomical time-series is challenging due to low signal-to-noise ratio, background star fields, and acquisition discontinuities across sequences. This paper presents a lightweight single-object asteroid tracing pipeline for FITS time-series that combines ZScale-based intensity normalization, frame differencing for motion segmentation, blob extraction using area-constrained detection, and greedy multi-frame association for track formation. Candidate tracks are filtered and ranked using a motion-focused scoring function that favors sufficient displacement, linearity (straightness), and temporal continuity. Experiments were conducted on approximately 4000 FITS frames containing multiple sequences, where frame-size changes are treated as sequence boundaries to avoid invalid differencing. The system outputs an interpretable trajectory overlay and quantitative motion statistics (path length, displacement, and average pixel speed). In a representative run, the selected object was tracked for 18 frames with 96.1 px net displacement and 9.7 px/frame average speed. The results suggest that combining simple difference imaging with heuristic track scoring can produce stable tracklets under noisy backgrounds while remaining computationally inexpensive compared to heavier tracking frameworks.