Track a small, distant drone in real time on low power hardware like a Raspberry Pi, no GPU required.
Run a local web demo to click-select and visually track a drone in an uploaded video clip.
Benchmark and tune a drone tracker's accuracy against annotated datasets using standard tracking metrics.
| punklabs-ai/womprat | 13127905/deep-learning-based-air-gesture-text-recognition- | 42kyynfqjv-dot/deepseekradio | |
|---|---|---|---|
| Stars | 15 | 15 | 15 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | moderate | hard |
| Complexity | 4/5 | 3/5 | 4/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Python 3.10+ and a virtual environment with OpenCV and FastAPI extras installed.
Womprat is a tool that watches a video feed and keeps track of a single drone as it moves across the frame, without needing a GPU. It is built for counter drone systems, the kind of setup used to spot and follow small unmanned aircraft that might be far away and only a few dozen pixels wide on screen, even when the sky is cluttered with birds, clouds, or a shaky camera. Once you give it a starting box around the target in the first frame, it locks on and keeps following that same object frame by frame, and it is designed to recover the lock if the drone briefly disappears behind something or the detector misses a few frames. Under the hood it combines several classical computer vision techniques rather than a deep learning model: a blob detector that spots moving shapes, a Kalman filter that predicts where the target should be next based on its motion, a matching step that connects detections to the right track, and some lightweight visual cues to help re-identify the target after it is briefly lost. This combination is tuned to run in real time on modest hardware, such as a quad core Raspberry Pi with no graphics card. Optional YOLO model weights are also included as an alternative detector, if you want to swap in a machine learning based detector instead of the classical one. The project includes a small web based demo, built with FastAPI, where you can upload or pick a video clip, click on the drone to lock onto it, and watch the tracking result overlaid on the video stream in your browser. This demo has no login protection, so it is meant to run only on your own computer or a trusted local network, not exposed to the internet. The repository also ships a full benchmarking system that measures tracking accuracy using standard metrics like ID switches and lost frames, tested against datasets of real drone and bird footage, so that any change to the tracker's tuning can be checked against measured results before being adopted. The code is released under the Apache License 2.0, though the datasets and any downloaded model weights used for testing are not covered by that license and have their own separate terms.
A CPU-only classical computer vision tracker that locks onto and follows a single small drone across video frames, no GPU needed.
Mainly Python. The stack also includes Python, OpenCV, FastAPI.
Apache 2.0, use freely including commercially, but datasets and model weights have their own separate terms.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.