print 'Training... ( N =', N, ')'
shuffler = cross_validation.ShuffleSplit(N, 1, 0.1)
for train_idx, test_idx in shuffler:
X_train = X[train_idx,:]
X_test = X[test_idx,:]
y_train = y[train_idx]
y_test = y[test_idx]
clf = svm.SVC(C=1.0, kernel='rbf', degree=3, gamma=0.003, verbose=False)
clf.fit(X_train, y_train)
pred = clf.predict(X_test)
print 'Correct rate:', or_ml.correctRate(pred, y_test)
gg = np.meshgrid(xrange(0, 180, 1), xrange(0, 256, 1))
dd = np.column_stack((gg[0].flat, gg[1].flat))
pred = clf.predict(dd)
pred2d = pred.reshape(256, 180)
joblib.dump(pred2d, opt.outfile)
classnames = map(or_util.removeExtension, filenames)
joblib.dump(classnames, or_util.appendToName(opt.outfile, '_classes'))
if opt.plot:
for i in xrange(0, C):
plt.plot(dd[pred==i,0], dd[pred==i,1], '.')
plt.axis([0, 180, 0, 255])
for hs in features:
plt.plot(hs[:,0], hs[:,1], 'o')
# plt.legend(filenames)
plt.show()
def __init__(self, opt):
self.woodenTable = joblib.load(opt.wooden)
self.woodenClasses = joblib.load(or_util.appendToName(opt.wooden, '_classes'))
self.plasticTable = joblib.load(opt.plastic)
self.plasticClasses = joblib.load(or_util.appendToName(opt.plastic, '_classes'))
self.materialClf = joblib.load(opt.material)
self.materialClasses = joblib.load(or_util.appendToName(opt.material, '_classes'))
self.purpleClf = joblib.load(opt.purple)
self.purpleClasses = joblib.load(or_util.appendToName(opt.purple, '_classes'))
rospy.init_node('object_recognition')
self.rate = rospy.Rate(LOOP_RATE)
self.blurTune = rospy.get_param('~blur_tune', False)
self.blobTune = rospy.get_param('~blob_tune', False)
self.cannyTune = rospy.get_param('~canny_tune', False)
self.morphTune = rospy.get_param('~morph_tune', False)
self.houghTune = rospy.get_param('~hough_tune', False)
self.show = rospy.get_param('~show', False)
self.tuneSetup = rospy.get_param('~tune_setup', False)
self.collect = rospy.get_param('~collect', 'nothing')
if self.collect != 'nothing':
rospy.on_shutdown(self.shutdown_hook)
if self.show:
if self.tuneSetup:
cv2.namedWindow(WINDOW_NAME, cv2.WINDOW_AUTOSIZE)
cv2.moveWindow(WINDOW_NAME, 400, 250)
cv2.namedWindow(TRACKBAR_WINDOW, cv2.WINDOW_AUTOSIZE)
cv2.moveWindow(TRACKBAR_WINDOW, 1050, 250)
else:
cv2.namedWindow(WINDOW_NAME, cv2.WINDOW_AUTOSIZE)
cv2.moveWindow(WINDOW_NAME, 600, 250)
if self.blurTune:
def blurSizeTune(v): self.blurSize = v
cv2.createTrackbar('Size', TRACKBAR_WINDOW, self.blurSize, 15, blurSizeTune)
def sigmaXTune(v): self.sigmaX = v
cv2.createTrackbar('Sigma x', TRACKBAR_WINDOW, self.sigmaX, 200, sigmaXTune)
if self.blobTune:
def minDistTune(v): self.minDist = v
cv2.createTrackbar('Min dist', TRACKBAR_WINDOW, self.minDist, 100, minDistTune)
def minAreaTune(v): self.minArea = v
cv2.createTrackbar('Min area', TRACKBAR_WINDOW, self.minArea, 2000, minAreaTune)
def maxAreaTune(v): self.maxArea = v
cv2.createTrackbar('Max area', TRACKBAR_WINDOW, self.maxArea, 40000, maxAreaTune)
def minCircTune(v): self.minCirc = v
cv2.createTrackbar('Min circ', TRACKBAR_WINDOW, self.minCirc, 100, minCircTune)
def minInerTune(v): self.minIner = v
cv2.createTrackbar('Min iner', TRACKBAR_WINDOW, self.minIner, 100, minInerTune)
if self.cannyTune:
def cannyLowTune(v): self.cannyLow = v
cv2.createTrackbar('Canny low', TRACKBAR_WINDOW, self.cannyLow, 1000, cannyLowTune)
def cannyHighTune(v): self.cannyHigh = v
cv2.createTrackbar('Canny high', TRACKBAR_WINDOW, self.cannyHigh, 1000, cannyHighTune)
def cannyKernelTune(v): self.cannyKernel = v
cv2.createTrackbar('Canny kernel', TRACKBAR_WINDOW, self.cannyKernel, 31, cannyKernelTune)
if self.morphTune:
def morphSizeTune(v): self.morphSize = v
cv2.createTrackbar('Morph size', TRACKBAR_WINDOW, self.morphSize, 31, morphSizeTune)
if self.houghTune:
def cannyParamTune(v): self.cannyParam = v
cv2.createTrackbar('Canny param', TRACKBAR_WINDOW, self.cannyParam, 200, cannyParamTune)
def param1Tune(v): self.param1 = v
cv2.createTrackbar('Param1', TRACKBAR_WINDOW, self.param1, 100, param1Tune)
self.blobParams = cv2.SimpleBlobDetector_Params()
self.blobParams.minDistBetweenBlobs = self.minDist
self.blobParams.filterByArea = True
self.blobParams.minArea = max(1, self.minArea)
self.blobParams.maxArea = max(self.minArea, self.maxArea)
self.blobParams.filterByColor = False
self.blobParams.filterByCircularity = True
self.blobParams.minCircularity = self.minCirc / 100.
self.blobParams.filterByConvexity = True
self.blobParams.minConvexity = self.minConv
self.blobParams.filterByInertia = True
self.blobParams.minInertiaRatio = self.minIner / 100.
# These 3 are not useful
# self.blobParams.minThreshold = 10
# self.blobParams.maxThreshold = 240
# self.blobParams.thresholdStep = max(1, self.stepThre)
def rgbCallback(rosimg): self.rgbRosImg = rosimg
self.rgbSub = rospy.Subscriber('/camera/rgb/image_raw', Image, rgbCallback)
def depthCallback(rosimg): self.depthRosImg = rosimg
self.depthSub = rospy.Subscriber('/camera/depth_registered/image_raw', Image, depthCallback)
self.recognitonProvider = rospy.Service(
'/object_recognition_provider',
Joint,
self.recognitionCallback)
self.tracker = or_ml.TrackerWrapper()
