void execVisualizerThread()
{
initVisualizer();
while(!visualizer_->wasStopped())
{
render(1);
}
visualizer_.reset();
}
PclCameraTrajectoryVisualizerImpl(bool render_thread = true)
{
if(render_thread)
{
visualizer_thread_.reset(new boost::thread(boost::bind(&PclCameraTrajectoryVisualizerImpl::execVisualizerThread, this)));
// wait for visualizer_ being created
while(!visualizer_)
{
boost::this_thread::yield();
}
}
else
{
initVisualizer();
}
}
int main(){
int i = 0;
int mapCount = 0, clearMapCount = 0, dumpCount=0;
int revFrameCount = 0;
#ifdef USE_NORTH
targetsGPS[maxTargets].lat = ADVANCED5LAT;
targetsGPS[maxTargets].lon = ADVANCED5LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED6LAT;
targetsGPS[maxTargets].lon = ADVANCED6LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED7LAT;
targetsGPS[maxTargets].lon = ADVANCED7LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED8LAT;
targetsGPS[maxTargets].lon = ADVANCED8LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED2LAT;
targetsGPS[maxTargets].lon = ADVANCED2LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED1LAT;
targetsGPS[maxTargets].lon = ADVANCED1LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED3LAT;
targetsGPS[maxTargets].lon = ADVANCED3LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED12LAT;
targetsGPS[maxTargets].lon = ADVANCED12LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED4LAT;
targetsGPS[maxTargets].lon = ADVANCED4LON;
maxTargets++;
#else
targetsGPS[maxTargets].lat = ADVANCED4LAT;
targetsGPS[maxTargets].lon = ADVANCED4LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED1LAT;
targetsGPS[maxTargets].lon = ADVANCED1LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED2LAT;
targetsGPS[maxTargets].lon = ADVANCED2LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED3LAT;
targetsGPS[maxTargets].lon = ADVANCED3LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED11LAT;
targetsGPS[maxTargets].lon = ADVANCED11LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED8LAT;
targetsGPS[maxTargets].lon = ADVANCED8LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED7LAT;
targetsGPS[maxTargets].lon = ADVANCED7LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED6LAT;
targetsGPS[maxTargets].lon = ADVANCED6LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED11LAT;
targetsGPS[maxTargets].lon = ADVANCED11LON;
maxTargets++;
targetsGPS[maxTargets].lat = ADVANCED5LAT;
targetsGPS[maxTargets].lon = ADVANCED5LON;
maxTargets++;
#endif
maxTargetIndex=maxTargets-1;
for(i=0;i<maxTargets;i++){// this is converting all GPS point data to XY data.
targetListXY[i].x = GPSX(targetsGPS[i].lon, startLongitude);
targetListXY[i].y = GPSY(targetsGPS[i].lat, startLatitude);
}
currentXY.x = GPSX(gpsvar.longitude,startLongitude);// converts current robot X location compared to start longitude
currentXY.y = GPSY(gpsvar.latitude,startLatitude);// converts current robot Y location compared to start latitude
targetXY = targetListXY[currentTargetIndex];//sets first target GPS point
nextTargetIndex = (currentTargetIndex + 1)%maxTargets;//sets next target GPS point
nextXY = targetListXY[nextTargetIndex];// ??
previousXY.x = GPSX(startLongitude, startLongitude);// why?
previousXY.y = GPSY(startLatitude, startLatitude);//Why?
initRoboteq(); /* Initialize roboteq */
initGuide();//what is guide?
#ifdef USE_VISION // if USE_vision is defined, then initialize vision.
initVision();
#endif //USE_VISION
#ifdef USE_GPS// if USE_GPS is defined, then initialize GPS.
initGPS();
initParser();
#endif //USE_GPS
#ifdef USE_LIDAR// if USE_LIDAR is defined, then initialize LIDAR.
initObjects();
initSICK();
#endif //USE_LIDAR
#ifdef DEBUG_VISUALIZER// if defined, then use visualizer.
initVisualizer();
#endif //DEBUG_VISUALIZER
#ifdef USE_MAP//////>>>>>>>>>>>????
initMap(0,0,0);
#endif //USE_MAP
#ifdef DUMP_GPS// dump GPS data into file
FILE *fp;
fp = fopen("gpsdump.txt", "w");
#endif // DUMP_GPS
while(1){
double dir = 1.0;
double speed = 0.0, turn = 0.0;
static double turnBoost = 0.750;//Multiplier for turn. Adjust to smooth jerky motions. Usually < 1.0
static int lSpeed = 0, rSpeed = 0;//Wheel Speed Variables
if (joystick() != 0) {// is joystick is connected
if (joy0.buttons & LB_BTN) {// deadman switch, but what does joy0.buttons do?????????????????????????????????
speed = -joy0.axis[1]; //Up is negative on joystick negate so positive when going forward
turn = joy0.axis[0];
lSpeed = (int)((speed + turnBoost*turn)*maxSpeed);//send left motor speed
rSpeed = (int)((speed - turnBoost*turn)*maxSpeed);//send right motor speed
}else{ //stop the robot
rSpeed=lSpeed=0;
}
if(((joy0.buttons & B_BTN)||autoOn)&& (saveImage==0)){//what is the single & ???????????????????
saveImage =DEBOUNCE_FOR_SAVE_IMAGE;//save each image the camera takes, save image is an int declared in vision_nav.h
}else{
if (saveImage) saveImage--; // turn off if button wasn't pressed?
}
if(joy0.buttons & RB_BTN){//turn on autonmous mode if start??? button is pressed
autoOn = 1;
mode=1;
}
if(joy0.buttons & Y_BTN){ // turn off autonomous mode
autoOn = 0;
mode =0;
}
lastButtons = joy0.buttons;//is this just updating buttons?
} else{
// printf("No Joystick Found!\n");
rSpeed=lSpeed=0;
}
//
// printf("3: %f %f\n",BASIC3LAT,BASIC3LON);
// printf("4: %f %f\n",BASIC4LAT,BASIC4LON);
// printf("5: %f %f\n",BASIC5LAT,BASIC5LON);
// getchar();
#ifdef AUTO_SWAP//what is this
if((currentTargetIndex>1&&targetIndexMem!=currentTargetIndex)||!autoOn||!mode==3){
startTime=currentTime=(float)(clock()/CLOCKS_PER_SEC);
targetIndexMem = currentTargetIndex;
}else{
currentTime=(float)(clock()/CLOCKS_PER_SEC);
}
totalTime = currentTime-startTime;
if(totalTime>=SWAPTIME&&autoOn){
swap();
targetIndexMem = 0;
}
#endif //AUTO_SWAP
#ifdef USE_GPS
readGPS();
currentXY.x = GPSX(gpsvar.longitude,startLongitude);
currentXY.y = GPSY(gpsvar.latitude,startLatitude);
robotTheta = ADJUST_RADIANS(DEG2RAD(gpsvar.course));
#else
currentXY.x = 0.0;
currentXY.y = 0.0;
robotTheta = 0.0;
#endif //USE_GPS
if(autoOn&&!flagPointSet){//this whole thing?????
flagXY.x=currentXY.x+FLAG_X_ADJUST;
flagXY.y=currentXY.y;
flagPointSet=1;
startAutoTime=currentAutoTime=(float)(clock()/CLOCKS_PER_SEC);
}
if(autoOn){
currentAutoTime=(float)(clock()/CLOCKS_PER_SEC);
totalAutoTime = currentAutoTime-startAutoTime;
if(totalAutoTime>=MODE2DELAY){
mode1TimeUp=1;//what is mode1 time up?
}
printf("TIMEING\n");
}
// if(currentTargetIndex <= OPEN_FIELD_INDEX || currentTargetIndex >= maxTargetIndex){
if(currentTargetIndex <= OPEN_FIELD_INDEX){//if you are on your last target, then set approaching thresh, and dest thresh to larger values?
//OPEN_FIELD_INDEX is set to 0 above...?
approachingThresh=4.0;
destinationThresh=3.0;
}else{//otherwise set your thresholds to a bit closer.
// destinationThresh=1.0;
destinationThresh=0.75;
approachingThresh=2.5;
}
//mode1 = lane tracking and obstacle avoidance. mode 2 = vision, lane tracking, but guide to gps. its not primary focus.
//mode3= gps mode in open field, but vision is toned down to not get distracted by random grass.
//mode 4= flag tracking
if(guide(currentXY, targetXY, previousXY, nextXY, robotTheta, robotWidth, 1)&& !allTargetsReached){//If target reached and and not all targets reached
printf("REACHED TARGET\n");
initGuide();// reset PID control stuff. problably resets all control variables.
previousXY = targetXY;//update last target
if(currentTargetIndex == maxTargetIndex){ //seeing if you are done with all targets.
allTargetsReached = 1;
}else{//otherwise update all the target information
currentTargetIndex = (currentTargetIndex + 1);
nextTargetIndex = (currentTargetIndex + 1)% maxTargets;
targetXY = targetListXY[currentTargetIndex];
nextXY = targetListXY[nextTargetIndex];
}
}
if((autoOn&&(currentTargetIndex == 0&&!approachingTarget&&!mode1TimeUp))||allTargetsReached){
//if autonomous, and on first target, and not not approaching target, and not mode 1 time up, or reached last target.
mode =1;//wtf is mode
distanceMultiplier = 50;//wthis is how heavily to rely on vision
} else if((autoOn&¤tTargetIndex == 0&&mode1TimeUp)||(autoOn&&approachingTarget&&(currentTargetIndex<=OPEN_FIELD_INDEX||currentTargetIndex>=maxTargetIndex-END_LANE_INDEX))){
mode =2;
distanceMultiplier = 50;
} else if((autoOn&¤tTargetIndex!=0)){
mode =3;
distanceMultiplier = 12;
}
flagPointDistance = D((currentXY.x-flagXY.x),(currentXY.y-flagXY.y));// basically the distance formula, but to what? what flags GPS point?
if(allTargetsReached&&flagPointDistance<FLAG_DIST_THRESH){
mode =4;// what is mode
}
#ifdef FLAG_TESTING
/*FLAG TESTING*/
mode=4;
#endif //FLAG_TESTING
/*Current Target Heading PID Control Adjustment*/
cvar.lookAhead = 0.00;//?
cvar.kP = 0.20; cvar.kI = 0.000; cvar.kD = 0.15;
turn = cvar.turn;
int bestVisGpsMask = 99;
int h = 0;
double minVisGpsTurn = 9999;
for(h=0;h<11;h++){
if(fabs((cvar.turn-turn_angle[h]))<minVisGpsTurn){
minVisGpsTurn=fabs((cvar.turn-turn_angle[h]));
bestVisGpsMask = h;
}
}
bestGpsMask = bestVisGpsMask;
// printf("bvg: %d \n", bestVisGpsMask);
// printf("vgt: %f cv3: %f\n", minVisGpsTurn,cvar3.turn);
#ifdef USE_VISION
// double visTurnBoost = 0.50;
double visTurnBoost = 1.0;
if(imageProc(mode) == -1) break;
if(mode==1||mode==2){
turn = turn_angle[bestmask];
turn *= visTurnBoost;
}else if(mode==3 && fabs(turn_angle[bestmask])>0.70){
turn = turn_angle[bestmask];
turn *= visTurnBoost;
}
#endif //USE_VISION
#ifdef USE_LIDAR
updateSick();
// findObjects();
#endif //USE_LIDAR
#ifdef USE_COMBINED_BUFFER//??????????
#define WORSTTHRESH 10
#define BESTTHRESH 3
if(mode==4){
#ifdef USE_NORTH
turn = (0.5*turn_angle[bestBlueMask]+0.5*turn_angle[bestRedMask]);
#else
turn = (0.65*turn_angle[bestBlueMask]+0.35*turn_angle[bestRedMask]);
#endif
turn *= 0.75;
}
combinedTargDist = cvar.targdist;
if(((approachingTarget||inLastTarget)&¤tTargetIndex>OPEN_FIELD_INDEX
&¤tTargetIndex<maxTargetIndex-END_LANE_INDEX)||(MAG(howbad[worstmask]-howbad[bestmask]))<BESTTHRESH||mode==4){
getCombinedBufferAngles(0,0);//Don't Use Vision Radar Data
}else{
getCombinedBufferAngles(0,1);//Use Vision Radar Data
}
if(combinedBufferAngles.left != 0 || combinedBufferAngles.right !=0){
if(mode == 1 || mode==2 || mode==3 || mode==4){
// if(mode == 1 || mode==2 || mode==3){
// if(mode==2 || mode==3){
// if(mode==3){
if(fabs(combinedBufferAngles.right)==fabs(combinedBufferAngles.left)){
double revTurn;
double revDistLeft, revDistRight;
int revIdx;
if(fabs(turn)<0.10) dir = -1.0;
if(fabs(combinedBufferAngles.left)>1.25) dir = -1.0;
if(dir<0){
revIdx = 540-RAD2DEG(combinedBufferAngles.left)*4;
revIdx = MIN(revIdx,1080);
revIdx = MAX(revIdx,0);
revDistLeft = LMSdata[revIdx];
revIdx = 540-RAD2DEG(combinedBufferAngles.right)*4;
revIdx = MIN(revIdx,1080);
revIdx = MAX(revIdx,0);
revDistRight = LMSdata[revIdx];
if(revDistLeft>=revDistRight){
revTurn = combinedBufferAngles.left;
}else {
revTurn = combinedBufferAngles.right;
}
turn = revTurn;
}else{
turn = turn_angle[bestmask];
}
} else if(fabs(combinedBufferAngles.right-turn)<fabs(combinedBufferAngles.left-turn)){
// } else if(turn<=0){
turn = combinedBufferAngles.right;
}else {
turn = combinedBufferAngles.left;
}
}
}
#endif //USE_COMBINED_BUFFER
if(dir<0||revFrameCount!=0){
dir = -1.0;
revFrameCount = (revFrameCount+1)%REVFRAMES;
}
// turn *= dir;
turn = SIGN(turn) * MIN(fabs(turn), 1.0);
speed = 1.0/(1.0+1.0*fabs(turn))*dir;
speed = SIGN(speed) * MIN(fabs(speed), 1.0);
if(!autoOn){
maxSpeed = 60;
targetIndexMem = 0;
}else if(dir<0){
maxSpeed = 30;
}else if(mode<=2||(mode==3 && fabs(turn_angle[bestmask])>0.25)){
maxSpeed = 60 - 25*fabs(turn);
// maxSpeed = 70 - 35*fabs(turn);
// maxSpeed = 90 - 50*fabs(turn);
// maxSpeed = 100 - 65*fabs(turn);
}else if(mode==4){
maxSpeed = 45-20*fabs(turn);
}else{
maxSpeed = 85 - 50*fabs(turn);
// maxSpeed = 100 - 65*fabs(turn);
// maxSpeed = 110 - 70*fabs(turn);
// maxSpeed = 120 - 85*fabs(turn);
}
if(autoOn){
lSpeed = (speed + turnBoost*turn) * maxSpeed;
rSpeed = (speed - turnBoost*turn) * maxSpeed;
}
#ifdef DEBUG_MAIN
printf("s:%.4f t: %.4f m: %d vt:%f dir:%f tmr: %f\n", speed, turn, mode, turn_angle[bestmask], flagPointDistance, totalAutoTime);
#endif //DEBUG_MAIN
#ifdef DUMP_GPS
if(dumpCount==0){
if (fp != NULL) {
fprintf(fp, "%f %f %f %f %f\n",gpsvar.latitude,gpsvar.longitude, gpsvar.course, gpsvar.speed, gpsvar.time);
}
}
dumpCount = dumpCount+1%DUMPGPSDELAY;
#endif //DUMP_GPS
#ifdef DEBUG_TARGET
debugTarget();
#endif //DEBUG_TARGET
#ifdef DEBUG_GUIDE
debugGuide();
#endif //DEBUG_GUIDE
#ifdef DEBUG_GPS
debugGPS();
#endif //DEBUG_GPS
#ifdef DEBUG_LIDAR
debugSICK();
#endif //DEBUG_LIDAR
#ifdef DEBUG_BUFFER
debugCombinedBufferAngles();
#endif //DEBUG_BUFFE
#ifdef DEBUG_VISUALIZER
robotX = currentXY.x;
robotY = currentXY.y;
robotTheta = robotTheta;//redundant I know....
targetX = targetXY.x;
targetY = targetXY.y;
// should probably pass the above to the function...
paintPathPlanner(robotX,robotY,robotTheta);
showPlot();
#endif //VISUALIZER
#ifdef USE_MAP
if(mapCount==0){
// mapRobot(currentXY.x,currentXY.y,robotTheta);
if(clearMapCount==0) clearMapSection(currentXY.x,currentXY.y,robotTheta);
else clearMapCount = (clearMapCount+1)%CLEARMAPDELAY;
mapVSICK(currentXY.x,currentXY.y,robotTheta);
// mapVSICK(0,0,0);
#ifdef USE_LIDAR
mapSICK(currentXY.x,currentXY.y,robotTheta);
#endif
showMap();
// printf("MAPPING\n");
}
mapCount= (mapCount+1)%MAPDELAY;
#endif //USE_MAP
sendSpeed(lSpeed,rSpeed);
Sleep(5);
}
#ifdef DUMP_GPS
fclose(fp);
#endif
return 0;
}
/** \brief The machine learning classifier, results are stored in the ClusterData structs.
* \param[in] cloud_in A pointer to the input point cloud.
* \param[in/out] clusters_data An array of information holders for each cluster
*/
void
applyObjectClassification (const pcl::PointCloud<PointType>::Ptr cloud_in, boost::shared_ptr<std::vector<ClusterData> > &clusters_data)
{
// Set up the machine learnin class
pcl::SVMTrain ml_svm_training; // To train the classifier
pcl::SVMClassify ml_svm_classify; // To classify
std::vector<pcl::SVMData> featuresSet; // Create the input vector for the SVM class
std::vector<std::vector<double> > predictionOut; // Prediction output vector
// If the input model_filename exists, it starts the classification.
// Otherwise it starts a new machine learning training.
if (global_data.model.size() > 0)
{
if (!ml_svm_classify.loadClassifierModel (global_data.model.data()))
return;
pcl::console::print_highlight (stderr, "Loaded ");
pcl::console::print_value (stderr, "%s ", global_data.model.data());
// Copy the input vector for the SVM classification
for (size_t c_it = 0; c_it < clusters_data->size (); ++c_it)
featuresSet.push_back ( (*clusters_data) [c_it].features);
ml_svm_classify.setInputTrainingSet (featuresSet); // Set input clusters set
ml_svm_classify.saveNormClassProblem ("data_input"); //Save clusters features
ml_svm_classify.setProbabilityEstimates (1); // Estimates the probabilities
ml_svm_classify.classification ();
}
else
{
// Currently: analyze on voxels of 0.08 x 0.08 x 0.08 meter with slight alteration based on cluster aggressiveness
float resolution = 0.08 * global_data.scale / pow (0.5 + global_data.cagg, 2);
// Create the viewer
pcl::visualization::PCLVisualizer viewer ("cluster viewer");
// Output classifier model name
global_data.model.assign (global_data.cloud_name.data());
global_data.model.append (".model");
std::vector<bool> lab_cluster;// save whether a cluster is labelled
std::vector<int> pt_clst_pos; // used to memorize in the total cloud, the point affiliation to the original cluster
// fill the vector (1 = labelled, 0 = unlabelled)
lab_cluster.resize ( (std::size_t) clusters_data->size ());
for (size_t c_it = 0; c_it < clusters_data->size (); ++c_it)
{
if ( (*clusters_data) [c_it].is_isolated)
{
(*clusters_data) [c_it].features.label = 0;
lab_cluster[c_it] = 1;
}
else
lab_cluster[c_it] = 0;
}
// Build a cloud with unlabelled clusters
pcl::PointCloud<PointType>::Ptr fragm_cloud (new pcl::PointCloud<PointType>);
// Initialize the viewer
initVisualizer (viewer);
PointType picked_point; // changed whether a mouse click occours. It saves the selected cluster index
viewer.registerPointPickingCallback (&pp_callback, (void *) &picked_point);
// Create a cloud with unlabelled clusters
for (size_t c_it = 0; c_it < clusters_data->size (); ++c_it)
if (!lab_cluster[c_it])
{
pcl::PointCloud<PointType>::Ptr cluster (new pcl::PointCloud<PointType>);
pcl::copyPointCloud (*cloud_in, * (*clusters_data) [c_it].indices, *cluster);
// Downsample cluster
pcl::VoxelGrid<PointType> sor;
sor.setInputCloud (cluster);
sor.setLeafSize (resolution, resolution, resolution);
sor.filter (*cluster);
// Copy cluster into a global cloud
fragm_cloud->operator+= (*cluster);
// Fill a vector to memorize the original affiliation of a point to the cluster
for (int clust_pt = 0; clust_pt < cluster->size(); clust_pt++)
pt_clst_pos.push_back (c_it);
// Add cluster to the viewer
std::stringstream cluster_name;
cluster_name << "cluster" << c_it;
pcl::visualization::PointCloudColorHandlerGenericField<PointType> rgb (cluster, "intensity");// Get color handler for the cluster cloud
viewer.addPointCloud<PointType> (cluster, rgb, cluster_name.str().data());
}
// Create a tree for point searching in the total cloud
pcl::KdTreeFLANN<pcl::PointXYZI> tree_;
tree_.setInputCloud (fragm_cloud);
// Visualize the whole cloud
int selected = -1; // save the picked cluster
bool stop = 0;
while (!viewer.wasStopped())
{
viewer.registerKeyboardCallback (keyboardEventOccurred, (void*) &stop);
boost::this_thread::sleep (boost::posix_time::microseconds (100000));
viewer.spinOnce (500);
if (picked_point.x != 0.0f || picked_point.y != 0.0f || picked_point.z != 0.0f) // if a point is clicked
{
std::vector<int> pointIdxNKNSearch (1);
std::vector<float> pointNKNSquaredDistance (1);
pcl::PointCloud<PointType>::Ptr cluster (new pcl::PointCloud<PointType>);
tree_.nearestKSearch (picked_point, 1, pointIdxNKNSearch, pointNKNSquaredDistance);
selected = pt_clst_pos[pointIdxNKNSearch[0]];
viewer.removePointCloud("cluster");
pcl::copyPointCloud (*cloud_in, * (*clusters_data) [selected].indices, *cluster);
pcl::visualization::PointCloudColorHandlerGenericField<PointType> rgb (cluster, "intensity");
viewer.addPointCloud<PointType> (cluster, rgb, "cluster");
picked_point.x = 0.0f;
picked_point.y = 0.0f;
picked_point.z = 0.0f;
}
if(selected != -1 && stop)
{
std::stringstream cluster_name;
cluster_name << "cluster" << selected;
lab_cluster[ selected ] = 1; // cluster is marked as labelled
(*clusters_data) [ selected ].features.label = 1; // the cluster is set as a noise
viewer.removePointCloud(cluster_name.str().data());
viewer.removePointCloud("cluster");
stop = 0;
selected = -1;
}
}
// Close the viewer
viewer.close();
// The remaining unlabelled clusters are marked as "good"
for (int c_it = 0; c_it < lab_cluster.size(); c_it++)
{
if (!lab_cluster[c_it])
(*clusters_data) [c_it].features.label = 0; // Mark remaining clusters as good
}
// Copy the input vector for the SVM classification
for (size_t c_it = 0; c_it < clusters_data->size (); ++c_it)
featuresSet.push_back ( (*clusters_data) [c_it].features);
// Setting the training classifier
pcl::SVMParam trainParam;
trainParam.probability = 1; // Estimates the probabilities
trainParam.C = 512; // Initial C value of the classifier
trainParam.gamma = 2; // Initial gamma value of the classifier
ml_svm_training.setInputTrainingSet (featuresSet); // Set input training set
ml_svm_training.setParameters (trainParam);
ml_svm_training.trainClassifier(); // Train the classifier
ml_svm_training.saveClassifierModel (global_data.model.data()); // Save classifier model
pcl::console::print_highlight (stderr, "Saved ");
pcl::console::print_value (stderr, "%s ", global_data.model.data());
ml_svm_training.saveTrainingSet ("data_input"); // Save clusters features normalized
// Test the current classification
ml_svm_classify.loadClassifierModel (global_data.model.data());
ml_svm_classify.setInputTrainingSet (featuresSet);
ml_svm_classify.setProbabilityEstimates (1);
ml_svm_classify.classificationTest ();
}
ml_svm_classify.saveClassificationResult ("prediction"); // save prediction in outputtext file
ml_svm_classify.getClassificationResult (predictionOut);
// Get labels order
std::vector<int> labels;
ml_svm_classify.getLabel (labels);
// Store the boolean output inside clusters_data
for (size_t c_it = 0; c_it < clusters_data->size (); ++c_it)
switch ( (int) predictionOut[c_it][0])
{
case 0:
(*clusters_data) [c_it].is_good = true;
break;
case 1:
(*clusters_data) [c_it].is_ghost = true;
break;
case 2:
(*clusters_data) [c_it].is_tree = true;
break;
}
// Store the percentage output inside cluster_data
for (size_t lab = 0; lab < labels.size (); lab++)
switch (labels[lab])
{
case 0:
for (size_t c_it = 0; c_it < clusters_data->size (); ++c_it)
{
(*clusters_data) [c_it].is_good_prob = predictionOut[c_it][lab + 1];
}
break;
case 1:
for (size_t c_it = 0; c_it < clusters_data->size (); ++c_it)
{
(*clusters_data) [c_it].is_ghost_prob = predictionOut[c_it][lab + 1];
}
break;
case 2:
for (size_t c_it = 0; c_it < clusters_data->size (); ++c_it)
{
(*clusters_data) [c_it].is_tree_prob = predictionOut[c_it][lab + 1];
}
break;
}
};
