Getting Trajectory objects from videos

There are several methods to discern the position of an object through consecutive frames of a video.

Color Matching

If the input is a video where the color of the object you want to track is quite different from everything else, like this one:

You can exploit this fact to capture the whole trajectory using the ColorMatching algorithm implemented in yupi:

from yupi.tracking import ColorMatching
algorithm = ColorMatching((180,125,35), (190,135,45))

Where the parameters passed to the ColorMatching constructor are the lower and upper bounds of the color vector in the selected color space, BGR by default.

Next, we can define a Region-of-Interest (ROI), the neighborhood of pixels, around the last known position of the object, that are going to be explored in the following frame. Its size will depend on the specific video and the desired tracking object.

from yupi.tracking import ROI
roi = ROI((100, 100))

Now, we can create an ObjectTracker. Its function is to apply the selected algorithm along the ROI in a frame to estimate the following position of the object.

from yupi.tracking import ObjectTracker
blue_ball = ObjectTracker('blue', algorithm, roi)

Finally, the tracker is passed to the TrackingScenario, the one in charge of iterating the video and making the trackers update its value on each frame. It also allows several trackers to coexist while processing the same video.

from yupi.tracking import TrackingScenario
scenario = TrackingScenario([blue_ball])

The track method of a TrackingScenario object, will produce a list of all the Trajectory objects the TrackingScenario tracked among all the frames of the video:

retval, tl = scenario.track('resources/videos/demo.avi', pix_per_m=10)

In this case, the list tl will contain only one object describing the trajectory of the blue ball in the video.

Other Tracking Algorithms

There are several other algorithms available in yupi (see Algorithms section on the API Reference). For a quick overview, we recommend you to inspect Example 6 that contains a detailed comparison of them while developing a specific task.

Tracking objects when the camera is moving

Yupi is able to estimate the motion of the camera (under certain circumstances) and integrate this information while reconstructing the trajectory of the tracked object. Example 5 shows a typical application of this feature.