This repository has been archived on 2020-04-08. You can view files and clone it, but cannot push or open issues or pull requests.
Files
YASMIN/nav/Filter.h

251 lines
7.7 KiB
C++

#ifndef FILTER_H
#define FILTER_H
#include <KLib/math/filter/particles/ParticleFilter.h>
#include <KLib/math/filter/particles/estimation/ParticleFilterEstimationWeightedAverage.h>
#include <KLib/math/filter/particles/estimation/ParticleFilterEstimationOrderedWeightedAverage.h>
#include <KLib/math/filter/particles/resampling/ParticleFilterResamplingSimple.h>
#include <KLib/math/filter/particles/resampling/ParticleFilterResamplingPercent.h>
#include <Indoor/sensors/radio/WiFiProbabilityFree.h>
#include <Indoor/sensors/radio/model/WiFiModelLogDistCeiling.h>
#include <Indoor/sensors/radio/WiFiProbabilityFree.h>
#include <Indoor/sensors/radio/WiFiProbabilityGrid.h>
#include <Indoor/grid/walk/v2/modules/WalkModuleHeadingControl.h>
#include <Indoor/grid/walk/v2/modules/WalkModuleNodeImportance.h>
#include <Indoor/grid/walk/v2/modules/WalkModuleFavorZ.h>
#include <Indoor/grid/walk/v2/modules/WalkModuleActivityControl.h>
#include <Indoor/grid/walk/v2/modules/WalkModuleFollowDestination.h>
#include "State.h"
#include "Node.h"
#include "NodeResampling.h"
#include "../Settings.h"
#include <omp.h>
#include <future>
class PFInit : public K::ParticleFilterInitializer<MyState> {
private:
Grid<MyGridNode>* grid;
public:
PFInit(Grid<MyGridNode>* grid) : grid(grid) {
}
virtual void initialize(std::vector<K::Particle<MyState>>& particles) override {
std::minstd_rand gen;
std::uniform_int_distribution<int> distIdx(0, grid->getNumNodes()-1);
std::uniform_real_distribution<float> distHead(0, 2*M_PI);
for (K::Particle<MyState>& p : particles) {
const int idx = distIdx(gen);
const MyGridNode& node = (*grid)[idx];
p.state.position = node; // random position
p.state.heading.direction = Heading(distHead(gen)); // random heading
p.weight = 1.0 / particles.size(); // equal weight
}
// // fix position + heading
// for (K::Particle<MyState>& p : particles) {
//// const int idx = 9000;
//// const MyGridNode& node = (*grid)[idx];
// const MyGridNode& node = grid->getNodeFor(GridPoint(2000, 2000, 0)); // center of the testmap
// p.state.position = node;
// p.state.heading.direction = Heading(0);
// }
}
};
class PFTrans : public K::ParticleFilterTransition<MyState, MyControl> {
public:
/** local, static control-data COPY */
MyControl ctrl;
Grid<MyGridNode>* grid;
GridWalker<MyGridNode, MyState> walker;
WalkModuleFavorZ<MyGridNode, MyState> modFavorZ;
WalkModuleHeadingControl<MyGridNode, MyState, MyControl> modHeading;
WalkModuleNodeImportance<MyGridNode, MyState> modImportance;
WalkModuleFollowDestination<MyGridNode, MyState> modDestination;
WalkModuleActivityControl<MyGridNode, MyState, MyControl> modActivity;
NodeResampling<MyState, MyGridNode> resampler;
std::minstd_rand gen;
public:
PFTrans(Grid<MyGridNode>* grid) : grid(grid), modHeading(&ctrl, Settings::IMU::turnSigma), modDestination(*grid), modActivity(&ctrl), resampler(*grid) {
//walker.addModule(&modFavorZ);
walker.addModule(&modHeading);
//walker.addModule(&modImportance);
walker.addModule(&modActivity);
if (Settings::destination != GridPoint(0,0,0)) {
//walker.addModule(&modDestination);
modDestination.setDestination(grid->getNodeFor(Settings::destination));
}
}
void transition(std::vector<K::Particle<MyState>>& particles, const MyControl* _ctrl) override {
// local copy!! observation might be changed async outside!! (will really produces crashes!)
this->ctrl = *_ctrl;
((MyControl*)_ctrl)->resetAfterTransition();
std::normal_distribution<float> noise(0, Settings::IMU::stepSigma);
// sanity check
Assert::equal((int)particles.size(), Settings::numParticles, "number of particles does not match the settings!");
//for (K::Particle<MyState>& p : particles) {
#pragma omp parallel for num_threads(3)
for (int i = 0; i < Settings::numParticles; ++i) {
//#pragma omp atomic
const float dist_m = std::abs(ctrl.numStepsSinceLastTransition * Settings::IMU::stepLength + noise(gen));
K::Particle<MyState>& p = particles[i];
double prob;
p.state = walker.getDestination(*grid, p.state, dist_m, prob);
//p.weight *= prob;//(prob > 0.01) ? (1.0) : (0.15);
//p.weight = (prob > 0.01) ? (1.0) : (0.15);
//p.weight = prob;
//p.weight = 1.0; // reset
//p.weight = std::pow(p.weight, 0.1); // make all particles a little more equal [less strict]
//p.weight *= std::pow(prob, 0.1); // add grid-walk-probability
p.weight = prob; // grid-walk-probability
if (p.weight != p.weight) {throw Exception("nan");}
}
}
};
class PFEval : public K::ParticleFilterEvaluation<MyState, MyObservation> {
Grid<MyGridNode>* grid;
WiFiModelLogDistCeiling& wifiModel;
//WiFiObserverFree wiFiProbability; // free-calculation
WiFiObserverGrid<MyGridNode> wiFiProbability; // grid-calculation
// how to perform VAP grouping. also see calibration in Controller.cpp
VAPGrouper vg = VAPGrouper(VAPGrouper::Mode::LAST_MAC_DIGIT_TO_ZERO, VAPGrouper::Aggregation::AVERAGE);
// smartphone is 1.3 meter above ground
const Point3 person = Point3(0,0,Settings::smartphoneAboveGround);
public:
PFEval(Grid<MyGridNode>* grid, WiFiModelLogDistCeiling& wifiModel) :
grid(grid), wifiModel(wifiModel),
//wiFiProbability(Settings::WiFiModel::sigma, wifiModel) { // WiFi free
wiFiProbability(Settings::WiFiModel::sigma) { // WiFi grid
}
double getStairProb(const K::Particle<MyState>& p, const ActivityButterPressure::Activity act) {
const float kappa = 0.75;
const MyGridNode& gn = grid->getNodeFor(p.state.position);
switch (act) {
case ActivityButterPressure::Activity::STAY:
if (gn.getType() == GridNode::TYPE_FLOOR) {return kappa;}
if (gn.getType() == GridNode::TYPE_DOOR) {return kappa;}
{return 1-kappa;}
case ActivityButterPressure::Activity::UP:
case ActivityButterPressure::Activity::DOWN:
if (gn.getType() == GridNode::TYPE_STAIR) {return kappa;}
if (gn.getType() == GridNode::TYPE_ELEVATOR) {return kappa;}
{return 1-kappa;}
}
return 1.0;
}
double evaluation(std::vector<K::Particle<MyState>>& particles, const MyObservation& _observation) override {
double sum = 0;
// local copy!! observation might be changed async outside!! (will really produces crashes!)
const MyObservation observation = _observation;
// vap-grouping
const int numAP1 = observation.wifi.entries.size();
const WiFiMeasurements wifiObs = vg.group(_observation.wifi);
const int numAP2 = wifiObs.entries.size();
Log::add("Filter", "VAP: " + std::to_string(numAP1) + " -> " + std::to_string(numAP2));
// sanity check
Assert::equal((int)particles.size(), Settings::numParticles, "number of particles does not match the settings!");
#pragma omp parallel for num_threads(3)
for (int i = 0; i < Settings::numParticles; ++i) {
K::Particle<MyState>& p = particles[i];
// WiFi free
//const double pWiFi = wiFiProbability.getProbability(p.state.position.inMeter()+person, observation.currentTime, vg.group(observation.wifi));
// WiFi grid
const MyGridNode& node = grid->getNodeFor(p.state.position);
const double pWiFi = wiFiProbability.getProbability(node, observation.currentTime, wifiObs);
//Log::add("xxx", std::to_string(observation.currentTime.ms()) + "_" + std::to_string(wifiObs.entries[0].ts.ms()));
const double pStair = getStairProb(p, observation.activity);
const double pGPS = 1;
const double prob = pWiFi * pGPS * pStair;
p.weight *= prob; // NOTE: keeps the weight returned by the transition step!
//p.weight = prob; // does NOT keep the weights returned by the transition step
if (p.weight != p.weight) {throw Exception("nan");}
#pragma omp atomic
sum += p.weight;
}
return sum;
}
};
#endif // FILTER_H