class KMeans(IClassifier):
def __init__(self, recorder=None):
self.name = 'kmeans'
self.recorder = recorder
self.kmeansGUI = False
self.currentRecordFrame = None
self.kmH = kmHelper.kMeansHepler()
self.cSignal = cSignal.SignalToGUI()
self.kmeans, self.kmeans_16N = None, None
self.bufferArray, self.bufferArray_16N = [], []
self.currentClass_16N = -3
self.firstCleanUpMessage = True
self.checkOnline = False
self.startBuffern = False
self.percentRatio = 0.0
self.wasClass3Before, self.wasClass4Before = False, False
self.classArray = None
self.recorder.classifyStart(self)
#cann't be start in the ui_kmenas_visualizer
if os.name == 'nt':
self.isWindows = True
self.shell = win32com.client.Dispatch("WScript.Shell")
else:
self.isWindows = False
#start the kmeans GUI
def startGui(self, recorder, callback):
self.kmeansGUI = True
self.app = ViewUIKMeans(self, self.getName, self.cSignal)
code = self.app.start()
callback(code)
def getName(self):
return self.name
def startTraining(self, args=[]):
pass
#classify online data
def classify(self, data):
if self.kmeansGUI:
cluster_0, cluster_1 = None, None
if self.startBuffern:
self.bufferArray = np.roll(self.bufferArray, -1, axis=0)
self.bufferArray[47] = self.kmH.normalizeSignalLevelSecond(data)
if self.checkOnline:
result = self.kmH.segmentOneHandGesture(self.bufferArray, outArrayLength=24, leftMargin=4, oneSecPeak=False)
if result is not None:
result = self.kmH.reduceDimensionality(result, setAxisTo=1, std='cut')
result = np.asarray(result)
if len(result) == self.kmeans.cluster_centers_.shape[1]:
cluster_0 = self.kMeansOnline(result)
if cluster_0 == 3:
if self.wasClass3Before:
print '------------------------------- by 3'
self.cSignal.emitSignal(10)
else:
if self.wasClass4Before:
#sometime came a single 3 after many 4's so this should fix it
self.wasClass4Before = False
else:
self.wasClass3Before = True
print ' ', cluster_0, '\n'
self.cSignal.emitSignal(10)
self.cSignal.emitSignal(3)
elif cluster_0 == 4:
if self.wasClass4Before:
print '------------------------------- by 4'
self.cSignal.emitSignal(10)
else:
self.wasClass4Before = True
print ' ', cluster_0,'\n'
self.cSignal.emitSignal(4)
self.cSignal.emitSignal(10)
else:
self.wasClass4Before = False
self.wasClass3Before = False
else:
print 'result length not matched !!!! : ', len(result)
if self.kmeans_16N is not None:
self.bufferArray_16N = np.roll(self.bufferArray_16N, -1, axis=0)
self.bufferArray_16N[57] = self.kmH.normalizeSignalLevelSecond(data)
if self.checkOnline:
result = self.kmH.segmentOneHandGesture(self.bufferArray[0:27,], outArrayLength=24, leftMargin=12, oneSecPeak=True)
if result is not None:
result = self.kmH.reduceDimensionality(result, setAxisTo=1, std='cut')
result = np.asarray(result)
if len(result) == self.kmeans_16N.cluster_centers_.shape[1]:
cluster_1 = self.kMeansOnline_16N(result)
if not self.wasClass4Before and not self.wasClass3Before:
if self.currentClass_16N != cluster_1:
#self.cSignal.emitSignal(cluster_1)
self.currentClass_16N = cluster_1
if cluster_1 == 0:
print cluster_1, '\n'
if self.isWindows:
self.shell.SendKeys("{PGUP}",0)
self.cSignal.emitSignal(0)
self.cSignal.emitSignal(10)
if cluster_1 == 1:
print ' ', cluster_1, '\n'
if self.isWindows:
self.shell.SendKeys("{PGDN}",0)
self.cSignal.emitSignal(1)
self.cSignal.emitSignal(10)
else:
if cluster_1 == 0:
print '------------------------------- by 0'
self.cSignal.emitSignal(10)
if cluster_1 == 1:
print '------------------------------- by 1'
self.cSignal.emitSignal(10)
else:
print 'result length not matched 16N !!!! : ', len(result)
if not self.wasClass4Before and not self.wasClass3Before and cluster_1 == 2 and self.firstCleanUpMessage:
self.firstCleanUpMessage = False
self.cSignal.emitSignal(-10)
else:
self.firstCleanUpMessage = True
def setClassArray(self, cArray):
self.classArray = cArray
def startValidation(self):
pass
def load(self, filename=""):
pass
def save(self, filename=""):
pass
def loadData(self, filename=""):
pass
def saveData(self, filename=""):
pass
def printClassifier(self):
pass
#initialize the two buffer for the two cluster and start to buffer
def fillBuffer(self, bufferSize):
self.bufferSize = bufferSize
self.bufferArray = np.zeros((48, 64))
self.bufferArray_16N = np.zeros((58, 64))
self.startBuffern = True
#set the two kmeans cluster initialized by ui_kmeans_visualizer
def setKMeans(self, kMeans, kMeans_16N=None):
self.kmeans = kMeans
self.kmeans_16N = kMeans_16N
#get the cluster assignment for current gesture and compare it with the
#class array we compute before
#we use here a class array because gesture three and four were performed twice
#in a soft and hard way and we compute a 5k cluster for them (the fifth gesture is the noise)
#by the online classification we put this two performance together
def kMeansOnline(self, checkArray):
class_ = self.kmeans.transform(checkArray)
cluster_ = self.kmH.checkClusterDistance(class_, self.percentRatio)
if self.classArray is not None:
clusterArray = np.where(self.classArray==cluster_)[0]
if clusterArray.size == 1:
cluster = clusterArray[0]
cluster_ = cluster
return cluster_
#get the cluster assignment for current gesture and returned it
#for gesture zero, one and noise we just use a 3k cluster
def kMeansOnline_16N(self, checkArray):
class_ = self.kmeans_16N.transform(checkArray)
cluster = self.kmH.checkClusterDistance(class_, self.percentRatio)
return cluster
#set online recognition
def checkKMeansOnline(self):
self.checkOnline = not self.checkOnline
def setPercentRatio(self, pRatio):
self.percentRatio = pRatio
class KMeans(IClassifier):
def __init__(self, recorder=None):
self.name = 'kmeans'
self.recorder = recorder
self.kmeansGUI = False
self.currentRecordFrame = None
self.kmH = kmHelper.kMeansHepler()
self.cSignal = cSignal.SignalToGUI()
self.kmeans, self.kmeans_16N = None, None
self.bufferArray, self.bufferArray_16N = [], []
self.currentClass_16N = -3
self.firstCleanUpMessage = True
self.checkOnline = False
self.startBuffern = False
self.percentRatio = 0.0
self.wasClass3Before, self.wasClass4Before = False, False
self.classArray = None
self.recorder.classifyStart(self)
#cann't be start in the ui_kmenas_visualizer
if os.name == 'nt':
self.isWindows = True
self.shell = win32com.client.Dispatch("WScript.Shell")
else:
self.isWindows = False
#start the kmeans GUI
def startGui(self, recorder, callback):
self.kmeansGUI = True
self.app = ViewUIKMeans(self, self.getName, self.cSignal)
code = self.app.start()
callback(code)
def getName(self):
return self.name
def startTraining(self, args=[]):
pass
#classify online data
def classify(self, data):
if self.kmeansGUI:
cluster_0, cluster_1 = None, None
if self.startBuffern:
self.bufferArray = np.roll(self.bufferArray, -1, axis=0)
self.bufferArray[47] = self.kmH.normalizeSignalLevelSecond(
data)
if self.checkOnline:
result = self.kmH.segmentOneHandGesture(self.bufferArray,
outArrayLength=24,
leftMargin=4,
oneSecPeak=False)
if result is not None:
result = self.kmH.reduceDimensionality(result,
setAxisTo=1,
std='cut')
result = np.asarray(result)
if len(result
) == self.kmeans.cluster_centers_.shape[1]:
cluster_0 = self.kMeansOnline(result)
if cluster_0 == 3:
if self.wasClass3Before:
print '------------------------------- by 3'
self.cSignal.emitSignal(10)
else:
if self.wasClass4Before:
#sometime came a single 3 after many 4's so this should fix it
self.wasClass4Before = False
else:
self.wasClass3Before = True
print ' ', cluster_0, '\n'
self.cSignal.emitSignal(10)
self.cSignal.emitSignal(3)
elif cluster_0 == 4:
if self.wasClass4Before:
print '------------------------------- by 4'
self.cSignal.emitSignal(10)
else:
self.wasClass4Before = True
print ' ', cluster_0, '\n'
self.cSignal.emitSignal(4)
self.cSignal.emitSignal(10)
else:
self.wasClass4Before = False
self.wasClass3Before = False
else:
print 'result length not matched !!!! : ', len(
result)
if self.kmeans_16N is not None:
self.bufferArray_16N = np.roll(self.bufferArray_16N,
-1,
axis=0)
self.bufferArray_16N[
57] = self.kmH.normalizeSignalLevelSecond(data)
if self.checkOnline:
result = self.kmH.segmentOneHandGesture(
self.bufferArray[0:27, ],
outArrayLength=24,
leftMargin=12,
oneSecPeak=True)
if result is not None:
result = self.kmH.reduceDimensionality(result,
setAxisTo=1,
std='cut')
result = np.asarray(result)
if len(
result
) == self.kmeans_16N.cluster_centers_.shape[1]:
cluster_1 = self.kMeansOnline_16N(result)
if not self.wasClass4Before and not self.wasClass3Before:
if self.currentClass_16N != cluster_1:
#self.cSignal.emitSignal(cluster_1)
self.currentClass_16N = cluster_1
if cluster_1 == 0:
print cluster_1, '\n'
if self.isWindows:
self.shell.SendKeys(
"{PGUP}", 0)
self.cSignal.emitSignal(0)
self.cSignal.emitSignal(10)
if cluster_1 == 1:
print ' ', cluster_1, '\n'
if self.isWindows:
self.shell.SendKeys(
"{PGDN}", 0)
self.cSignal.emitSignal(1)
self.cSignal.emitSignal(10)
else:
if cluster_1 == 0:
print '------------------------------- by 0'
self.cSignal.emitSignal(10)
if cluster_1 == 1:
print '------------------------------- by 1'
self.cSignal.emitSignal(10)
else:
print 'result length not matched 16N !!!! : ', len(
result)
if not self.wasClass4Before and not self.wasClass3Before and cluster_1 == 2 and self.firstCleanUpMessage:
self.firstCleanUpMessage = False
self.cSignal.emitSignal(-10)
else:
self.firstCleanUpMessage = True
def setClassArray(self, cArray):
self.classArray = cArray
def startValidation(self):
pass
def load(self, filename=""):
pass
def save(self, filename=""):
pass
def loadData(self, filename=""):
pass
def saveData(self, filename=""):
pass
def printClassifier(self):
pass
#initialize the two buffer for the two cluster and start to buffer
def fillBuffer(self, bufferSize):
self.bufferSize = bufferSize
self.bufferArray = np.zeros((48, 64))
self.bufferArray_16N = np.zeros((58, 64))
self.startBuffern = True
#set the two kmeans cluster initialized by ui_kmeans_visualizer
def setKMeans(self, kMeans, kMeans_16N=None):
self.kmeans = kMeans
self.kmeans_16N = kMeans_16N
#get the cluster assignment for current gesture and compare it with the
#class array we compute before
#we use here a class array because gesture three and four were performed twice
#in a soft and hard way and we compute a 5k cluster for them (the fifth gesture is the noise)
#by the online classification we put this two performance together
def kMeansOnline(self, checkArray):
class_ = self.kmeans.transform(checkArray)
cluster_ = self.kmH.checkClusterDistance(class_, self.percentRatio)
if self.classArray is not None:
clusterArray = np.where(self.classArray == cluster_)[0]
if clusterArray.size == 1:
cluster = clusterArray[0]
cluster_ = cluster
return cluster_
#get the cluster assignment for current gesture and returned it
#for gesture zero, one and noise we just use a 3k cluster
def kMeansOnline_16N(self, checkArray):
class_ = self.kmeans_16N.transform(checkArray)
cluster = self.kmH.checkClusterDistance(class_, self.percentRatio)
return cluster
#set online recognition
def checkKMeansOnline(self):
self.checkOnline = not self.checkOnline
def setPercentRatio(self, pRatio):
self.percentRatio = pRatio
def startGui(self, recorder, callback):
self.kmeansGUI = True
self.app = ViewUIKMeans(self, self.getName, self.cSignal)
code = self.app.start()
callback(code)
def startGui(self, recorder, callback):
self.kmeansGUI = True
self.app = ViewUIKMeans(self, self.getName, self.cSignal)
code = self.app.start()
callback(code)
