commit 4b4fe83c91e2a734cb518ccc57ba9735f143cd73
parent 2e83234987c9ff509019993a557831dc87781154
Author: Stefan Koch <taikano@lavabit.com>
Date: Tue, 11 Jun 2013 10:55:56 +0200
added training on keypress
Diffstat:
| M | models.py | | | 30 | ++---------------------------- |
| M | track2.py | | | 75 | +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++---------------- |
2 files changed, 61 insertions(+), 44 deletions(-)
diff --git a/models.py b/models.py
@@ -17,7 +17,7 @@ gestures = [
# TODO: Zuordnung von Indizes stimmt noch nicht
def transition_matrix(gesture):
gestes = list(set(gesture))
- A = [[0.01 for i in range(GESTES_COUNT)] for j in range(GESTES_COUNT)]
+ A = [[0.1 for i in range(GESTES_COUNT)] for j in range(GESTES_COUNT)]
# self transitions are high
for geste in gestes:
A[geste][geste] = 0.7
@@ -35,7 +35,7 @@ def transition_matrix(gesture):
def emission_matrix(gesture):
gestes = set(gesture)
- B = [[0.01 for i in range(OBSERVATIONS_COUNT)] for j in range(GESTES_COUNT)]
+ B = [[0.1 for i in range(OBSERVATIONS_COUNT)] for j in range(GESTES_COUNT)]
for geste in gestes:
if geste == UP:
@@ -67,29 +67,3 @@ for gesture in gestures:
m = ghmm.HMMFromMatrices(sigma, ghmm.DiscreteDistribution(sigma), A, B, pi)
print(m)
models.append((m, gesture[1]))
-
-close = models[0][0]
-close.baumWelch(ghmm.SequenceSet(sigma, [
- [0, 0, 0, 0, 3, 3, 3, 3, 3],
- [0, 0, 0, 0, 3, 3, 3, 3, 3, 1],
- [0, 0, 0, 0, 0, 3, 3, 3, 1],
- [0, 0, 0, 3, 3, 3],
- [0, 0, 0, 3, 3, 3, 1]
-]))
-
-print(close)
-
-#A = [[0.7, 0.3], [0, 1]]
-#B = [[0.6, 0.1, 0.2, 0.2], [.2, .2, .1, .6]]
-#pi = [1, 0]
-#m = ghmm.HMMFromMatrices(sigma, ghmm.DiscreteDistribution(sigma), A, B, pi)
-
-#A2 = [[0.7, 0.3], [0, 1]]
-#B2 = [[0.6, 0.1, 0.2, 0.2], [0.2, .2, .6, .1]]
-#pi2 = [1, 0]
-#m2 = ghmm.HMMFromMatrices(sigma, ghmm.DiscreteDistribution(sigma), A2, B2, pi2)
-
-#models = [
-# (m, ['/usr/bin/xdotool', 'getactivewindow', 'windowkill']),
-# (m2, ['notify-send', '"received command"', '"Hooray!"']),
-#]
diff --git a/track2.py b/track2.py
@@ -1,3 +1,5 @@
+#!/usr/bin/python2.7
+
import cv2
import math
import subprocess
@@ -29,6 +31,23 @@ def pointer_pos(img):
return (None, None)
+def movement_direction(x_delta, y_delta):
+ if abs(x_delta) > 10 or abs(y_delta) > 10:
+ degree = math.atan2(y_delta, x_delta)
+ if -0.75 * math.pi <= degree < -0.25 * math.pi:
+ direction = UP
+ elif -0.25 * math.pi <= degree < 0.25 * math.pi:
+ direction = LEFT
+ elif 0.25 * math.pi <= degree < 0.75 * math.pi:
+ direction = DOWN
+ else:
+ direction = RIGHT
+
+ return direction
+ else:
+ return None
+
+
def execute(emission_seq, models):
max_comm = None
max_val = 0
@@ -40,18 +59,26 @@ def execute(emission_seq, models):
max_val = res[1][-1]
max_comm = command
- if max_val >= 0.4:
+ if max_val >= 0.3:
subprocess.call(max_comm)
print(max_comm)
+def train(emission_seq, model):
+ model.baumWelch(emission_seq)
+
+
+# Two variables to determine changes in train-mode
+train_mode_pre = False
+train_mode = False
+train_target = 0
+
cam = cv2.VideoCapture(0)
winName = "Movement Indicator"
cv2.namedWindow(winName, cv2.CV_WINDOW_AUTOSIZE)
img = cv2.cvtColor(cam.read()[1], cv2.COLOR_BGR2HSV)
-cv2.imwrite("test.jpg", img)
img = cv2.inRange(img, (70, 100, 100), (150, 255, 255))
img = cv2.erode(img, numpy.array([[1,1,1],[1,1,1],[1,1,1]]))
img = cv2.dilate(img, numpy.array([[1,1,1],[1,1,1],[1,1,1]]), iterations=3)
@@ -61,6 +88,12 @@ x1, y1 = pointer_pos(img)
not_changed = 0
while True:
+ # if we switched to train mode, delete all prior knowledge
+ if train_mode_pre == False and train_mode == True:
+ path = []
+ not_changed = 0
+
+ train_mode_pre = train_mode
x0 = x1
y0 = y1
@@ -74,18 +107,9 @@ while True:
x_delta = x1 - x0
y_delta = y1 - y0
- if abs(x_delta) > 10 or abs(y_delta) > 10:
- degree = math.atan2(y_delta, x_delta)
- if -0.75 * math.pi <= degree < -0.25 * math.pi:
- path.append(UP)
- elif -0.25 * math.pi <= degree < 0.25 * math.pi:
- path.append(LEFT)
- elif 0.25 * math.pi <= degree < 0.75 * math.pi:
- path.append(DOWN)
- else:
- path.append(RIGHT)
- #print("Appended to")
- #print(path)
+ direction = movement_direction(x_delta, y_delta)
+ if direction is not None:
+ path.append(direction)
else:
not_changed += 1
else:
@@ -93,7 +117,14 @@ while True:
if not_changed > 5:
if len(path) >= 2:
print(path)
- execute(ghmm.EmissionSequence(models.sigma, path), models.models)
+ if train_mode == False:
+ execute(ghmm.EmissionSequence(models.sigma, path), models.models)
+ else:
+ print("Training model %d" % (train_target,))
+ train(ghmm.EmissionSequence(models.sigma, path), models.models[train_target][0])
+ train_mode = False
+ print("Leaving training mode")
+ print(models.models[train_target][0])
path = []
not_changed = 0
@@ -103,5 +134,17 @@ while True:
if key == 27:
cv2.destroyWindow(winName)
break
-
+ elif key == ord('0'):
+ train_mode = True
+ train_target = 0
+ elif key == ord('1'):
+ train_mode = True
+ train_target = 1
+ elif key == ord('2'):
+ train_mode = True
+ train_target = 2
+ elif key == ord('3'):
+ train_mode = True
+ train_target = 3
+
print "Goodbye"
