Calculate accuracy and distance that each prediction is wrong by (where 0 is it being correct)

Print out accuracy and the average wrong by distance.

Fix profile for set-1. The threshold needed to be a bit tighter due to the wooden table containing similar color values

Author: addisonElliott

algorithm.py

@@ -5,6 +5,10 @@
 
 
 def runAlgorithm(dir):
+    numCorrect = 0
+    numWrong = 0
+    wrongByDistance = list()
+
     # Loop through each file in directory
     for name in os.listdir(dir):
         fullPath = os.path.join(dir, name)
@@ -39,9 +43,18 @@ def runAlgorithm(dir):
                 print('Found filename that does not start with a valid number: %s' % (name2))
                 continue
 
-            print('yay %s' % fullPath2)
             image = scipy.misc.imread(fullPath2)
             predictedFingerCount = detectFingerCount(image, colorProfile)
 
-            # TODO Check if we got it right. Also, store statistics and analyze how we did
-            print('Should be: %i    Got: %i' % (actualFingerCount, predictedFingerCount))
+            print('Should be: %i    Got: %i' % (actualFingerCount, predictedFingerCount))
+            if predictedFingerCount == actualFingerCount:
+                numCorrect = numCorrect + 1
+            else:
+                numWrong = numWrong + 1
+
+                wrongByDistance.append(np.abs(predictedFingerCount - actualFingerCount))
+
+    accuracy = numCorrect / (numCorrect + numWrong)
+    averageWrongBy = sum(wrongByDistance) / len(wrongByDistance)
+    print('Algorithm Accuracy: %f%%' % (accuracy * 100))
+    print('Average Algorithm Wrong By: %f' % (averageWrongBy))

data/training/set-1/profile.xml

@@ -1,3 +1,3 @@
 <profile SchemaVersion="1">
-  <ycbcr minY="0" maxY="255" minCb="77" maxCb="127" minCr="145" maxCr="180" />
+  <ycbcr minY="120" maxY="255" minCb="77" maxCb="127" minCr="140" maxCr="180" />
 </profile>

detectFingerCount.py

@@ -44,6 +44,9 @@ def detectFingerCount(image, colorProfile):
     # Apply threshold based on color profile
     imageMask = thresholdYCbcr(imageYCbCr, colorProfile[0, :], colorProfile[1, :])
 
+    # # Testing image mask
+    # imageMask = thresholdYCbcr(imageYCbCr, np.array([120, 77, 140]), np.array([255, 127, 180]))
+
     # # Show YCbCr image and final mask
     # plt.figure(1)
     # plt.clf()
@@ -172,4 +175,4 @@ def detectFingerCount(image, colorProfile):
     # plt.waitforbuttonpress()
 
     # Number of fingers up is webs plus one
-    return fingerWebs + 1
+    return min(fingerWebs + 1, 5)