class MotionDetector():
"""Motion Detector class, gets images and performs a regression"""
DETECTION_TIME = .7 # Linear regression time
MOVING_WINDOW_LENGTH = 40 # Number of samples for regression window
AVERAGE_WINDOW_LENGTH = 20 # Number of samples in moving average window
MINIMUM_REGRESSION_SAMPLES = 5 # Samples skipped before starting regression
MEAN_THRESHOLD = 0.3 # Minimum change in image that will trigger detection
DEVIATION_THRESHOLD = 10 # Maximum deviation from linear behaviour accepted
Y_THRESHOLD = 90 # Maximum deviation from linear behaviour accepted
TRIGGER_TIME_THRESHOLD = 5 # Timeout between detection, in seconds
SPEED_THRESHOLD = 115 # Minimum speed that will trigger detection, in px/sec
REF_IM_UPDATE_TIME = 3 # time where little change happens until reference image is updated
REF_IM_WINDOW_LENGTH = 200
REF_IM_THRESHOLD = 30000
def __init__(self,camera):
self.regression_window = MovingWindow(self.MOVING_WINDOW_LENGTH) #Start a moving window for the regression
self.time_change_window = MovingWindow(self.REF_IM_WINDOW_LENGTH)
self.change_window = np.zeros(self.AVERAGE_WINDOW_LENGTH) # Start the moving average window for the motion detections
self.last_trigger_time = 0
self.camera = camera
def update(self,frame,diff_frame):
"""Update the detection status with a new image"""
self.gray_frame = np.uint8(np.mean(diff_frame,2))
self.image_moments = moments(self.gray_frame)
# Update the moving average window
self.change_window = np.roll(self.change_window,1)
self.change_window[0] = self.image_moments['m00']
self.mom_x = np.uint(self.image_moments['m10']/(self.image_moments['m00']+1))
self.mom_y = np.uint(self.image_moments['m01']/(self.image_moments['m00']+1))
# Detection window output
#self.gray_frame[self.mom_y:self.mom_y+5,self.mom_x:self.mom_x+5] = 255
#cv2.imshow('preview',self.gray_frame)
# Update the moving regression window with the new measurements
self.regression_window.push(self.mom_x)
self.time_change_window.push(self.image_moments['m00'])
# Update regressor variables
self.t_out,self.p_out,self.n_out = self.regression_window.fetch(self.DETECTION_TIME)
_,self.change_out,_ = self.time_change_window.fetch(self.REF_IM_UPDATE_TIME)
if np.max(self.change_out) < self.REF_IM_THRESHOLD:
self.camera.update_reference_frame(frame)
print np.max(self.change_out)
if len(self.t_out) > self.MINIMUM_REGRESSION_SAMPLES:
self.slope,self.intercept ,_,_,self.std_dev = stats.linregress(self.t_out,self.p_out)
############################################################
############### Dirty hack! Flipped webcam #################
## self.slope = -self.slope
############################################################
############################################################
self.pred_y = self.slope*self.t_out + self.intercept*np.ones(self.t_out.shape)
self.y_dev = np.sqrt(np.sum((self.p_out-self.pred_y)**2))
#print(self.y_dev)
def detection(self):
"""True is a detection has occurred at least TRIGGER_TIME_THRESHOLD ago"""
if len(self.t_out) >5:
#if self.change_window.mean() > self.MEAN_THRESHOLD and self.y_dev < self.Y_THRESHOLD and time() - self.last_trigger_time > self.TRIGGER_TIME_THRESHOLD:
if self.change_window.mean() > self.MEAN_THRESHOLD and self.std_dev < self.DEVIATION_THRESHOLD and time() - self.last_trigger_time > self.TRIGGER_TIME_THRESHOLD:
if self.slope > self.SPEED_THRESHOLD:
self.last_trigger_time = time()
return True
return False
def data(self):
"""Return regression data"""
if len(self.t_out) > self.MINIMUM_REGRESSION_SAMPLES:
return self.slope, self.intercept, self.y_dev, self.std_dev
else:
return 0,0,0,0
def dump(self):
return self.t_out, self.p_out