This video demonstrates multiple target tracking (MTT) using the MTT algorithm Recursive Ransac (R-RANSAC). The camera on the UAV acts as the sensor which feeds into R-RANSAC point measurements. These measurements are give to R-RANSAC to initialize tracks and maintain tracks. This implementation of R-RANSAC assumes constant velocity motion models. When a track is initialized and persists for a while, it is elevated to the status of a "good track". The good tracks are shown as a dot surrounded by a colored ellipse. The ellipse's color is used to distinguish one track from another. The ellipse also represents the validation region of the track. The validation region is a volume around the track that any measurement inside the validation region is used to update the track.
In this video, there are five people walking/running different trajectories: straight lines, curves, ellipses, etc. You can see that the R-RANSAC is able to track them fairly well, but loses the tracks 13 times for a short time. You will also notice tracks on the shadows. This is because the visual frontend that gives R-RANSAC measurements, is producing measurements from the shadows. So this artifact is not a result of R-RANSAC, but rather the front end.
