This project is an implementation for single object tracking based on mean-shift algorithm, coded in Python language.
Here we use 4 different videos (relatively easy to track) to show the tracking result, most of which are from www.pexels.com, an open community for sharing best free stock videos.
- Runnable Python source files are SingleObjectTracking.py and SingleObjectTracking_Built_in.py, and the former is implemented on our own, while the later is based on builtin mean-shift of OpenCV. JUST CLONE THE REPOSITORY AND RUN IT!
- Dataset is in dataset directory, which contains four sets of frames which are bird, girl_in_the_sea, girl_in_the_alley and girl_in_the_garden.
- Single object tracking results are in output directory, so that you can quickly get the result of this code.
- Calculate the probability histogram of the given target
- Calculate the probability histogram of the candidate target
- Find the candidate target which is the closest to the given target with mean-shift
- [1] D. Comaniciu and P. Meer, "Mean shift: a robust approach toward feature space analysis," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, no. 5, pp. 603-619, May 2002, doi: 10.1109/34.1000236.
- [2] D. Comaniciu, V. Ramesh and P. Meer, "Kernel-based object tracking," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 25, no. 5, pp. 564-577, May 2003, doi: 10.1109/TPAMI.2003.1195991.
- [3] http://cvlab.hanyang.ac.kr/tracker_benchmark/datasets.html
LeoHao (XMU-CS)
2020.11.20