Algorithm 1. The Pseudo-code of the Hierarchical Activity Recognition Method

Inputs:

(1) dataset with activity labels

(2) the hierarchical tree T associated with the activity labels

Output:

(1) a set of classifiers and feature subsets

1. cls_fs = {}; //a set used for storing outputs

2. For each non-leaf node nd do

(2.1) search the child nodes ↓(nd) of nd with T;

(2.2) obtain a training set Tr(nd) using ↓(nd);

(2.3) do feature selection on Tr(nd), and get Snd;

(2.4) construct a classifier clsnd using Tr(nd) and Snd;

(2.5) cls_fs.add(< clsnd, Snd>); // add it to cls_fs

3. return cls_fs