train_kf.m

%% train_kf.m
%  
% Script that trains the Kalman filter process and measurement noise parameters. Training images that are used are 
% extracted via get_training_data.py. 
%
% Inputs:
%   - Folder containing Kalman filter training images, extracted using the get_training_data.py script
%   - Ground truth pose estimates of the Kalman filter training images, labeled with label_images.m
%   - Trained CNN model for the subject
%
% Outputs:
%   - Q/R process and measurement noise parameters (respectively), saved in ./parameters/
%
% Translational Neuroengineering Laboratory (TNEL) @ UC San Diego
% Website: http://www.tnel.ucsd.edu

clear; close all;
dateTime = datestr(now,'mm-dd-yy_HH:MM:SS');

%% Setup & Options
run patientpose_setup
run patientpose_options

%% Load Kalman Filter Training Images
disp('Select the folder containing Kalman filter training images');
im.folder = uigetdir('','Folder containing Kalman filter training images');
addpath(im.folder);
im.files = dir(fullfile(im.folder,'*.jpg'));
im.names = natsortfiles({im.files.name});

%% Load Ground Truth Kalman Filter Training Poses
disp('Load the ground truth data for the Kalman filter training images');
uiopen('matlab');
kf_gt = detections.manual.locs;

%% Apply Model to Kalman Filter Training Data
opt.modelFile = './caffe-heatmap/models/heatmap-flic-fusion/caffe-heatmap-flic.caffemodel';
[pose_caffe, ~, ~, ~] = applyNet(im.names, opt, pp);

%% Train Kalman Filter Noise Parameters
[Q, R] = kf_train(pose_gt, pose_caffe);

%% Save
save([pwd '/parameters/QR_' dateTime '.mat'],'Q','R');