Part I: Wink detection Write an OpenCV program that can detect a winking face.

You may want to build your program by changing the example program DetectWink.py.

example program:

def detectWink(frame, location, ROI, cascade): eyes = cascade.detectMultiScale( ROI, 1.15, 3, 0|cv2.CASCADE_SCALE_IMAGE, (10, 20)) for e in eyes: e[0] += location[0] e[1] += location[1] x, y, w, h = e[0], e[1], e[2], e[3] cv2.rectangle(frame, (x,y), (x+w,y+h), (0, 0, 255), 2) return len(eyes) == 1 # number of eyes is one def detect(frame, faceCascade, eyesCascade): gray_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY) # possible frame pre-processing: # gray_frame = cv2.equalizeHist(gray_frame) # gray_frame = cv2.medianBlur(gray_frame, 5) scaleFactor = 1.15 # range is from 1 to .. minNeighbors = 3 # range is from 0 to .. flag = 0|cv2.CASCADE_SCALE_IMAGE # either 0 or 0|cv2.CASCADE_SCALE_IMAGE minSize = (30,30) # range is from (0,0) to .. faces = faceCascade.detectMultiScale( gray_frame, scaleFactor, minNeighbors, flag, minSize) detected = 0 for f in faces: x, y, w, h = f[0], f[1], f[2], f[3] faceROI = gray_frame[y:y+h, x:x+w] if detectWink(frame, (x, y), faceROI, eyesCascade): detected += 1 cv2.rectangle(frame, (x,y), (x+w,y+h), (255, 0, 0), 2) else: cv2.rectangle(frame, (x,y), (x+w,y+h), (0, 255, 0), 2) return detected def run_on_folder(cascade1, cascade2, folder): if(folder[-1] != "/"): folder = folder + "/" files = [join(folder,f) for f in listdir(folder) if isfile(join(folder,f))] windowName = None totalCount = 0 for f in files: img = cv2.imread(f, 1) if type(img) is np.ndarray: lCnt = detect(img, cascade1, cascade2) totalCount += lCnt if windowName != None: cv2.destroyWindow(windowName) windowName = f cv2.namedWindow(windowName, cv2.WINDOW_AUTOSIZE) cv2.imshow(windowName, img) cv2.waitKey(0) return totalCount def runonVideo(face_cascade, eyes_cascade): videocapture = cv2.VideoCapture(0) if not videocapture.isOpened(): print("Can't open default video camera!") exit() windowName = "Live Video" showlive = True while(showlive): ret, frame = videocapture.read() if not ret: print("Can't capture frame") exit() detect(frame, face_cascade, eyes_cascade) cv2.imshow(windowName, frame) if cv2.waitKey(30) >= 0: showlive = False # outside the while loop videocapture.release() cv2.destroyAllWindows() if __name__ == "__main__": # check command line arguments: nothing or a folderpath if len(sys.argv) != 1 and len(sys.argv) != 2: print(sys.argv[0] + ": got " + len(sys.argv) - 1 + "arguments. Expecting 0 or 1:[image-folder]") exit() # load pretrained cascades face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml') eye_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_eye.xml') if(len(sys.argv) == 2): # one argument folderName = sys.argv[1] detections = run_on_folder(face_cascade, eye_cascade, folderName) print("Total of ", detections, "detections") else: # no arguments runonVideo(face_cascade, eye_cascade) 

Suggestion of changes

1. Changing the parameters of the functions

2. Using different cascades

3. Including or changing different preprocessing steps

4. Changing region of interest

5. Changing the logic of the program.

6. Anything else which would improve the correct detection and reduce the incorrect detections.