Simeple Blue Circle Tracker is a simple program to track any blue circle in a video. This program has two parts: (1) blue circle detection and (2) blue circle tracking. This code shows a basic idea on multiple object tracking using an appearance model. You can find my demo here.
This algorithm aims to link blobs detected in the current frame with the tracks from the previous frame using a bounding box overlap rule. The first step is to construct an overlap area association matrix between the bounding box of blobs and tracks. This matrix gives us four possible cases: (1) existing object detected, (2) new object detected, (3) merging object detected, and (4) splitting object detected. However, the current program does not handle case (3) and (4). Also, it does not remove any track when it is no longer visible.
For case (1), the matrix will have at most one non-zero element in each row or column. It means each track associates with each blob, and each blob associates with each track, respectively. So that we can easily link blobs with tracks. For case (2), columns with all zero elements represent newly appeared objects in a video. It means these blobs do not associate with any existing track. So that we create new tracks for them.
This code does not handle any occlusion, so it would be great if this feature is implemented. It can also be improved by a better appearance model, Kalman filtering, or particle filtering.
This work is licensed under a Creative Commons Attribution 3.0 License. You are free to use the code here for any purpose, but please give credit to its source. If you make improvements, I will appreciate it if you send me your updates.