Structure Preserving Object Tracker


The Structure Preserving Object Tracker (SPOT) is able to track (i.e. to identify) multiple, arbitrary objects simultaneously in a video sequence based on a single (bounding-box) annotation of the objects of interest. Our tracker can be used for, e.g., surveillance, image stabilization, and monitoring of production processes. Its key advantage over alternative trackers is that it exploits structural relations between the objects during tracking.

SPOT was developed by Lu Zhang and Laurens van der Maaten of the Vision Lab at TU Delft. It runs in real-time, and achieves state-of-the-art performance on a range of standard benchmark videos. The tracker is described in the following papers:

  • L. Zhang and L.J.P. van der Maaten. Preserving Structure in Model-Free Tracking. IEEE Transactions on Pattern Recognition and Machine Intelligence, 36(4):756-769, 2014. [ PDF ]
  • L. Zhang and L.J.P. van der Maaten. Structure Preserving Object Tracking. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1838-1845, 2013. [ PDF, Slides ]

Source code

A simple Matlab implementation of the SPOT tracker can be obtained from here. The code contains a demo on the red flowers movie. The other annotated videos we used in our experiments are available as well: air show, basketball, car chase, hunting, parade, shaking, skating, and sky diving.

Please cite our work if you use our code or data!

For questions about our tracker, please contact Lu Zhang and/or Laurens van der Maaten.


The eight movies below demonstrate the performance of the SPOT Tracker. In some videos, the box on the right shows the current model of the object, as learned by the tracker.