This is a repo for course project of EL2320 Applied Estimation at KTH. This project is an implementation of an integrated face tracker using color-based and moment-based particle filters (PFs).

To test the functionality of the face tracker, videos of moving faces are fed into the system and particles as prediction are labeled in each frame of the video. Particles are in the represention of bounding boxes with different colors. The code is mainly done in Matlab. For more details, please read the report.

• Obtain initial state of the face target using Haar feature-based cascade classifier
• Generate particles around initial state and propagate them
• Measure histogram similarities between particles and face target for color-based and moment-based model
• Perform systematic resampling for particles
• Fuse estimations from color-based and moment-based model

• In the "code" folder, run the script main_ICM.m to see how the integrated face tracker works; run the script main_clr.m to see how the color-based face tracker works alone.
• In the "videos" folder, sample3.mp4 and sample4.mp4 are input videos, and the others are outputs from the face tracking system.
• In the "doc" folder, you can find the report with more details.

Red bounding boxes represent particles; blue bounding box represents posterior state.

Red and blue bounding boxes represent particles and posterior state from color-based model; yellow and black bounding boxes represent particles and posterior state from color-based model.

• Moment-based PF performs better dynamically and responds quickly when target changes its moving direction;
• Color-based PF is more stable but responds relatively slowly compared with moment-based PF;
• Integrated PF takes advantages of the above two PFs and outperforms them in general.

[1] Nummiaro, Katja, Esther Koller-Meier, and Luc Van Gool. "An adaptive color-based particle filter." Image and vision computing 21.1 (2003): 99-110. [paper]

[2] Junxiang, Gao, Zhou Tong, and Liu Yong. "Face tracking using color histograms and moment invariants." Broadband Network & Multimedia Technology, 2009. IC-BNMT'09. 2nd IEEE International Conference on. IEEE, 2009. [paper]