# Motion-Based Multiple Object Tracking [closed]

See also my reduced question: Tracing 2d coordinates by using Nearest function

A very important issue in many scientific fields such as Physics, Biology, Medicine and Astronomy is Motion-Based Multiple Object Tracking from videos or sequential images.

The following two steps have to be performed:

1. Detecting moving objects in each frame
2. Associating the detections corresponding to the same object over time

In any given frame, some detections may be assigned to tracks, while other detections remain unassigned. The assigned tracks are updated using the corresponding detections. On the other hand if newly unassigned detections occur they start new tracks.

Each object track keeps count of the number of consecutive frames, where it remained assigned and also stores required information about the the tracked object (size, shape, ...).

For the tracking often a motion dependent prediction is used to reduce the ROI where the corresponding same object can be detected.

In languages such as Matlab, Python and IDL, a lot of code has been written and can be freely downloaded from the internet. Is there some activity corresponding to this important issue in Mathematica?

I asked this to Wolfram Technical Support and the answer was: "There are examples in the reference page for the function ImageFeatureTrack".

How could ImageFeatureTrack be applied to detected morphological components to track e.g. their brightness centre from image to image?

• One of the problems I foresee is first choosing an appropriate algorithm for tracking object, and how to deal with missing or occluded objects in certain frames. This excellent review article in Nature Methods shows that different algorithms are appropriate for different scenarios. – dr.blochwave Dec 14 '15 at 9:54
• – Sjoerd C. de Vries Dec 14 '15 at 10:11