if you're using opencv2.4, then it's bad news: the KalmanFilter is unusable, since you cannot set the transition (or any other) Matrix.
for opencv3.0 it works correctly, like this:
import cv2, numpy as np
meas=[]
pred=[]
frame = np.zeros((400,400,3), np.uint8) # drawing canvas
mp = np.array((2,1), np.float32) # measurement
tp = np.zeros((2,1), np.float32) # tracked / prediction
def onmouse(k,x,y,s,p):
global mp,meas
mp = np.array([[np.float32(x)],[np.float32(y)]])
meas.append((x,y))
def paint():
global frame,meas,pred
for i in range(len(meas)-1): cv2.line(frame,meas[i],meas[i+1],(0,100,0))
for i in range(len(pred)-1): cv2.line(frame,pred[i],pred[i+1],(0,0,200))
def reset():
global meas,pred,frame
meas=[]
pred=[]
frame = np.zeros((400,400,3), np.uint8)
cv2.namedWindow("kalman")
cv2.setMouseCallback("kalman",onmouse);
kalman = cv2.KalmanFilter(4,2)
kalman.measurementMatrix = np.array([[1,0,0,0],[0,1,0,0]],np.float32)
kalman.transitionMatrix = np.array([[1,0,1,0],[0,1,0,1],[0,0,1,0],[0,0,0,1]],np.float32)
kalman.processNoiseCov = np.array([[1,0,0,0],[0,1,0,0],[0,0,1,0],[0,0,0,1]],np.float32) * 0.03
#kalman.measurementNoiseCov = np.array([[1,0],[0,1]],np.float32) * 0.00003
while True:
kalman.correct(mp)
tp = kalman.predict()
pred.append((int(tp[0]),int(tp[1])))
paint()
cv2.imshow("kalman",frame)
k = cv2.waitKey(30) &0xFF
if k == 27: break
if k == 32: reset()