This repository contains the Matlab code of my dissertation for the MSc in Artificial Intelligence at the University of Edinburgh.
A complete set of results can be seen in the following video:
https://www.youtube.com/watch?v=RZFBXTjE8Ik&t=1s
The directory named DEMO contains a small sample of the final results obtained, where you can inspect the point cloud of the person and the skeleton, fitted using particle filters.
In order to run the program, you will need to download this file (286Mb) and add it to the DEMO directory.
More detail instructions can be found inside the script DEMO.m, but you just basically need to run all the lines inside it.
The project is divided into three parts:
- pre-processing
- detection
- tracking
The main script of the project is main.m. This script calls the rest of the scripts.
Inside the main.m script, two main scripts are called:
main_create_pcl.mmain_particle_filter.m
The first one handles the pre-processing stage and the detection part. The second one takes care of the tracking part.
main_create_pcl.m calls the following relevant scripts:
personDetectionAllFrames.m: detects the person using color (blue gown and pink gloves) and depth (region growing algorithm)personDetectionAllFramesNoColorBack.m: detects the person using only depth (background subtraction method)
main_particle_filter.m calls the following relevant scripts:
particle_filter_rshoulder.mparticle_filter_relbow.mparticle_filter_lshoulder.mparticle_filter_lelbow.m
These four scripts contain the particle filters for each joint to be tracked. They could be joined as one in future improvements.
The rest of the scripts are called by one or more of the scripts described here.