/*
* Create a new person from the provided stamped pose and point cloud, and insert them into the map.
*/
Person& PersonDetector::create_person(const geometry_msgs::Pose& pose, const sensor_msgs::PointCloud2& cloud) {
// Make a new person with the trivial person classifier, for now, and then give them an initial pose.
Person lifeform(_current_uid++);
lifeform.push_pose(pose);
// boost::shared_ptr<PersonClassifier> classifier(new ShirtColorPersonClassifier(cloud, 30.0));
boost::shared_ptr<PersonClassifier> classifier(new TrivialPersonClassifier);
// Then insert them.
tracked()[lifeform.uid()] = lifeform;
classifiers()[lifeform.uid()] = classifier;
// Return a reference to the person in the map.
return tracked()[lifeform.uid()];
}
void learn(const database<double, bool> &data, const ublas::vector<double> &sample_weights){
ASSERT(data.patterns.size2() == data.targets.size(), " ");
ASSERT(is_equal(sum(sample_weights), 1.0, 1e-2), " ");
std::vector<tsf_classifier> classifiers(data.patterns.size1());
for (size_t i = 0; i < classifiers.size(); ++i){
classifiers[i] = _tsfc_prototype.clone();
classifiers[i].feature_index(i);
classifiers[i].learn(data, sample_weights);
}
auto min_it = std::min_element(classifiers.begin(), classifiers.end(),
[&](const tsf_classifier &lhs, const tsf_classifier &rhs){
return lhs.error() < rhs.error();
});
_best_classifier = classifiers[min_it - classifiers.begin()];
}
/*
* Attempt to find a classifier by UID. Returns a reference to the classifier if found, or an empty optional otherwise.
*/
boost::optional<PersonClassifier&> PersonDetector::classifier_by_id(unsigned int uid) {
if(classifiers().count(uid) >= 1)
return *classifiers()[uid];
return boost::none;
}
