Code for human pose representation followed by activity recognition using SVM hree approaches for pose repesentation have been implemented namely- Histogram of Oriented Displacements(HOD), Histogram of Joint Position Differences (HJPD), Relative Distances and Angles of Star Skeleton(RAD)
- C++11
This c++ code can be compiled through Linux terminal.
Open a new command window and cd to directory to Skeleton_code run the command
g++ -std=c++11 HOD.cpp to run HOD code. To check the output ./a.out run the command g++ -std=c++11 RAD.cpp to run RAD code. To check the output ./a.out run the command g++ -std=c++11 HJPD.cpp to run HJPD code. To check the output ./a.out
The output vectors are stored in
RAD
TEST rad_h1.t TRAIN rad_h1
HJPD
TEST hjpd_h1.t TRAIN hjpd_h1
HOD
TEST hod_h1.t TRAIN hod_h1
For the RAD configuration
The joints are taken as //centre = 1 //head = 4 //larm = 8 //rarm = 12 //lleg = 16 //rleg = 20 . The distances and angles are calculated from the first joint to the others. The bin sizes are taken as 10 for both angle and distance histograms. The code contains lines for implementing the training and test sets seperately. Uncommenting and commenting give both files. Vector length = 5*(M+N) = 5*(10+10) = 100
For the HJPD configuration
All the joints are taken into a factor and relative distances from the base joint (1) are calculated. The histogram bin size is taken as 5.The code contains lines for implementing the training and test sets seperately. Uncommenting and commenting give both files.
For the HOD configuration
Every joint is taken into consideration for the weighted histogram representation. Instead of adding one for the frequency the distance weight is added to the histogram