Changes, hopeflly fixes + debug tools

calibrate-capture.py

@@ -0,0 +1,20 @@
+import cv2
+
+c = cv2.VideoCapture(2)
+# if not ding this we only have  jittery 10fps
+c.set(cv2.CAP_PROP_FOURCC, cv2.VideoWriter_fourcc(*"MJPG"))
+
+# set camera resoltion
+c.set(3, 1280)
+c.set(4, 720)
+
+for i in range(15):
+    _,im = c.read()
+    cv2.imwrite('calibrate/left-{:06d}.png'.format(i), im)
+    cv2.imshow('left', im)
+    
+    if cv2.waitKey(1000) & 0xFF == ord('q'):
+        break
+    
+c.release()
+cv2.destroyAllWindows()

calibrate.py

@@ -0,0 +1,44 @@
+import numpy as np
+import cv2
+import glob
+
+# termination criteria
+criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 50, 0.001)
+# criteria=cv2.CALIB_CB_FAST_CHECK
+
+# prepare object points, like (0,0,0), (1,0,0), (2,0,0) ....,(6,5,0)
+objp = np.zeros((6*7,3), np.float32)
+objp[:,:2] = np.mgrid[0:7,0:6].T.reshape(-1,2)
+
+# Arrays to store object points and image points from all the images.
+objpoints = [] # 3d point in real world space
+imgpoints = [] # 2d points in image plane.
+
+images = glob.glob('calibrate/*.png')
+
+for fname in images:
+    print(fname)
+    img = cv2.imread(fname)
+    gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
+#     gray = cv2.resize(gray, (640, 360))
+
+    # Find the chess board corners
+    ret, corners = cv2.findChessboardCorners(gray, (3,3),None)
+    print(ret)
+
+    # If found, add object points, image points (after refining them)
+    if ret == True:
+        objpoints.append(objp)
+
+        corners2 = cv2.cornerSubPix(gray,corners,(11,11),(-1,-1),criteria)
+        imgpoints.append(corners2)
+
+        # Draw and display the corners
+        img = cv2.drawChessboardCorners(gray, (7,6), corners2,ret)
+        cv2.imshow('img',img)
+        cv2.waitKey(5000)
+
+cv2.destroyAllWindows()
+
+ret, mtx, dist, rvecs, tvecs = cv2.calibrateCamera(objpoints, imgpoints, gray.shape[::-1],None,None)
+print(ret, mtx, dist, rvecs, tvecs)

get_coordinates.py

@@ -0,0 +1,131 @@
+import pickle
+import numpy as np
+
+metricsSize = [960,600]
+screenDrawCorners = np.array([[0,0], [metricsSize[0]-1,0], [0, metricsSize[1]-1], [metricsSize[0]-1,metricsSize[1]-1]])
+
+def create_perspective_transform_matrix(src, dst):
+    """ Creates a perspective transformation matrix which transforms points
+        in quadrilateral ``src`` to the corresponding points on quadrilateral
+        ``dst``.
+
+        Will raise a ``np.linalg.LinAlgError`` on invalid input.
+        """
+    # See:
+    # * http://xenia.media.mit.edu/~cwren/interpolator/
+    # * http://stackoverflow.com/a/14178717/71522
+    in_matrix = []
+    for (x, y), (X, Y) in zip(src, dst):
+        in_matrix.extend([
+            [x, y, 1, 0, 0, 0, -X * x, -X * y],
+            [0, 0, 0, x, y, 1, -Y * x, -Y * y],
+        ])
+
+    A = np.matrix(in_matrix, dtype=np.float)
+    B = np.array(dst).reshape(8)
+    af = np.dot(np.linalg.inv(A.T * A) * A.T, B)
+    m = np.append(np.array(af).reshape(8), 1).reshape((3, 3))
+    return m
+
+# got this amazing thing from here: https://stackoverflow.com/a/24088499
+def create_perspective_transform(src, dst, round=False, splat_args=False):
+    """ Returns a function which will transform points in quadrilateral
+        ``src`` to the corresponding points on quadrilateral ``dst``::
+
+            >>> transform = create_perspective_transform(
+            ...     [(0, 0), (10, 0), (10, 10), (0, 10)],
+            ...     [(50, 50), (100, 50), (100, 100), (50, 100)],
+            ... )
+            >>> transform((5, 5))
+            (74.99999999999639, 74.999999999999957)
+
+        If ``round`` is ``True`` then points will be rounded to the nearest
+        integer and integer values will be returned.
+
+            >>> transform = create_perspective_transform(
+            ...     [(0, 0), (10, 0), (10, 10), (0, 10)],
+            ...     [(50, 50), (100, 50), (100, 100), (50, 100)],
+            ...     round=True,
+            ... )
+            >>> transform((5, 5))
+            (75, 75)
+
+        If ``splat_args`` is ``True`` the function will accept two arguments
+        instead of a tuple.
+
+            >>> transform = create_perspective_transform(
+            ...     [(0, 0), (10, 0), (10, 10), (0, 10)],
+            ...     [(50, 50), (100, 50), (100, 100), (50, 100)],
+            ...     splat_args=True,
+            ... )
+            >>> transform(5, 5)
+            (74.99999999999639, 74.999999999999957)
+
+        If the input values yield an invalid transformation matrix an identity
+        function will be returned and the ``error`` attribute will be set to a
+        description of the error::
+
+            >>> tranform = create_perspective_transform(
+            ...     np.zeros((4, 2)),
+            ...     np.zeros((4, 2)),
+            ... )
+            >>> transform((5, 5))
+            (5.0, 5.0)
+            >>> transform.error
+            'invalid input quads (...): Singular matrix
+        """
+    try:
+        transform_matrix = create_perspective_transform_matrix(src, dst)
+        error = None
+    except np.linalg.LinAlgError as e:
+        transform_matrix = np.identity(3, dtype=np.float)
+        error = "invalid input quads (%s and %s): %s" %(src, dst, e)
+        error = error.replace("\n", "")
+
+    to_eval = "def perspective_transform(%s):\n" %(
+        splat_args and "*pt" or "pt",
+    )
+    to_eval += "  res = np.dot(transform_matrix, ((pt[0], ), (pt[1], ), (1, )))\n"
+    to_eval += "  res = res / res[2]\n"
+    if round:
+        to_eval += "  return (int(round(res[0][0])), int(round(res[1][0])))\n"
+    else:
+        to_eval += "  return (res[0][0], res[1][0])\n"
+    locals = {
+        "transform_matrix": transform_matrix,
+    }
+    locals.update(globals())
+    exec to_eval in locals, locals
+    res = locals["perspective_transform"]
+    res.matrix = transform_matrix
+    res.error = error
+    return res
+
+def coordinatesToSrc(coordinates):
+    return np.array([coordinates['tl'], coordinates['tr'],coordinates['bl'], coordinates['br']])
+
+
+
+print("Coordinates in pickle file:")
+with open('coordinates.p', 'r') as fp:
+    c = pickle.load(fp)
+    for name, coord in c.items():
+        print("\t", name, coord[0], coord[1])
+    
+transform = create_perspective_transform(coordinatesToSrc(c), screenDrawCorners, True)
+
+print("Metrics")
+print(metricsSize)
+
+print("Test halfway point:")
+x = ((c['tl'][0]+c['bl'][0])/2 + (c['tr'][0]+c['br'][0])/2) / 2
+y = ((c['tl'][1]+c['tr'][1])/2 + (c['bl'][1]+c['br'][1])/2) / 2
+print("\t",x,y)
+print(transform((x,y)))
+
+print("tl", transform((c['tl'])))
+print("tr", transform((c['tr'])))
+print("bl", transform((c['bl'])))
+print("br", transform((c['br'])))
+        
+        

head_pose.py

@@ -76,6 +76,11 @@ argParser.add_argument(
     default=4,
     help="Nr of total processes (min 3)"
 )
+argParser.add_argument(
+    '--only-metrics',
+    action="store_true",
+    help="Render only metrics instead of the heatmap. Convenient for debugging."
+)
 
 args = argParser.parse_args()
 
@@ -88,10 +93,7 @@ logger = logging.getLogger(__name__)
 
 # im = cv2.imread("headPose.jpg");
 
-spotSize = (100,100)
+
-spot = Image.open(os.path.join(cur_dir,"spot.png")).convert('L')
-spot = spot.resize(spotSize)
-spot = np.array(spot)
 
 
 predictor_path = os.path.join(cur_dir,"shape_predictor_68_face_landmarks.dat")
@@ -106,11 +108,28 @@ screenDrawCorners = np.array([[10,60], [90, 60], [10, 110], [90, 110]])
 
 # metrics matrix
 metricsSize = [1920,1080]
-metricsSize = [1280,800]
+# metricsSize = [1280,800]
-metricsSize = [960,600]
+# metricsSize = [960,600]
+metricsSize = [1080,1080] # no point in having it different from to the render size
 dataframe = pd.DataFrame(columns=['x','y'])
 
 renderSize = [1280,800]
+renderSize = [1080,1080]
+
+# Used to create a black backdrop, instead of the ugly Qt-gray, if neccessary 
+screenSize = [1920,1080]
+
+spotS = int(100./720*renderSize[1])
+spotSize = (spotS, spotS)
+
+spot = Image.open(os.path.join(cur_dir,"spot.png")).convert('L')
+spot = spot.resize(spotSize)
+spot = np.array(spot)
+
+backdrop = None
+if screenSize != renderSize:
+    shape = [screenSize[1],screenSize[0], 3]
+    backdrop = np.zeros(shape, dtype=np.uint8)
 
 metrics = None
 if lastMetricsFilename and os.path.isfile(lastMetricsFilename):
@@ -231,7 +250,7 @@ def coordinatesToSrc(coordinates):
 # coordinates of the screen boundaries
 if os.path.exists("coordinates.p"):
     coordinates = pickle.load(open("coordinates.p", "rb"))
-    transform = create_perspective_transform(coordinatesToSrc(coordinates), screenDrawCorners)
+    transform = create_perspective_transform(coordinatesToSrc(coordinates), screenDrawCorners, True)
 
     a = [np.array([ 1312.15541183]), np.array([ 244.56278002]), 0]
     logger.info("Loaded coordinates: %s", coordinatesToSrc(coordinates))
@@ -306,6 +325,7 @@ def captureFacesPoints(i):
 
         # We use this later for calibrating
         currentPoint = None
+        currentVectors = None
         currentPoints = []
 
         if len(dets) > 0:
@@ -420,13 +440,14 @@ def captureFacesPoints(i):
                 # x = translation_vector[0] + rotation_vector[0]* a
                 # y = translation_vector[1] + rotation_vector[1] * a
                 # logger.warn("First {} {},{}".format(a,x,y))
-                a = - translation_vector[2]# / viewDirectionVector[2]
+                a = translation_vector[2] / viewDirectionVector[2]
                 x = translation_vector[0] + viewDirectionVector[0] * a
                 y = translation_vector[1] + viewDirectionVector[1] * a
                 # logger.warn("Second {} {},{}".format(a,x,y))
                 point = np.array([x,y])
 
                 currentPoint = point
+                currentVectors = {'translation': translation_vector, 'rotation': viewDirectionVector}
                 currentPoints.append(point)
 
                 td3 = time.time()
@@ -434,7 +455,7 @@ def captureFacesPoints(i):
 
                 # TODO only draw nose line now, so we can change color depending whether on screen or not
 
-        results = {'currentPoint': currentPoint, 'currentPoints': currentPoints}
+        results = {'currentPoint': currentPoint, 'currentPoints': currentPoints, 'currentVectors': currentVectors}
         results['im'] = im if not args.hide_preview else None
 
         try:
@@ -445,11 +466,14 @@ def captureFacesPoints(i):
 
 def captureVideo():
     c = cv2.VideoCapture(args.camera)
+    # if not ding this we only have  jittery 10fps
+    c.set(cv2.CAP_PROP_FOURCC, cv2.VideoWriter_fourcc(*"MJPG"))
+    
     # set camera resoltion
-    # c.set(3, 1280)
+    c.set(3, 1280)
-    # c.set(4, 720)
+    c.set(4, 720)
-    c.set(3, 960)
+#     c.set(3, 960)
-    c.set(4, 540)
+#     c.set(4, 540)
     logger.debug("Camera FPS: {}".format(c.get(5)))
 
     while True:
@@ -462,33 +486,36 @@ def captureVideo():
         logger.debug("Que sizes: image: {}, points: {} ".format(photoQueue.qsize(), pointsQueue.qsize()))
 
 
-processes = []
+if __name__ == '__main__':
-for i in range(args.processes - 2):
+    processes = []
+    for i in range(args.processes - 2):
         p = multiprocessing.Process(target=captureFacesPoints, args=(i,))
         p.daemon = True
         p.start()
         processes.append(p)
     
-p = multiprocessing.Process(target=captureVideo, args=())
+    p = multiprocessing.Process(target=captureVideo, args=())
-p.daemon = True
+    p.daemon = True
-p.start()
+    p.start()
-processes.append(p)
+    processes.append(p)
     
-newMetrics = np.zeros((metricsSize[1], metricsSize[0]))
+    newMetrics = np.zeros((metricsSize[1], metricsSize[0]))
-lastRunTime = 0
+    lastRunTime = 0
     
-while True:
+    while True:
         result = None
-    try:
         te1 = time.time()
+        try:
             result = pointsQueue.get()
             te1b = time.time()
             im = result['im']
             currentPoint = result['currentPoint']
             currentPoints = result['currentPoints']
+            currentVectors = result['currentVectors']
         except queue.Empty as e:
             logger.warn('Result queue empty')
     
+        tr1 = time.time()
         if result is not None:
             if not args.hide_preview:
                 # draw little floorplan for 10 -> 50, sideplan 60 -> 100 (40x40 px)
@@ -528,9 +555,13 @@ while True:
                     targetInt = (int(targetPoint[0]), int(targetPoint[1]))
                     # check if point fits on screen:
                     # if so, measure it
-                if targetInt[0]+spotSize[0] >= 0 and targetInt[1]+spotSize[1] >= 0 and targetInt[0]-spotSize[0] < metricsSize[0] and targetInt[1]-spotSize[0] < metricsSize[1]:
+                    if targetInt[0]+spotSize[0] >= 0 and targetInt[1]+spotSize[1] >= 0 and targetInt[0]-spotSize[0] < metricsSize[0] and targetInt[1]-spotSize[1] < metricsSize[1]:
+                        if not args.hide_graph:
                             dataframe = dataframe.append({'x':targetInt[0],'y':targetInt[1]}, ignore_index=True)
-                    logger.info("Put metric {},{} in metrix of {},{}".format(targetInt[1],targetInt[0], metricsSize[1], metricsSize[0]))
+                            
+                        logger.info("Put metric {},{} in metrics of {},{}".format(targetInt[1],targetInt[0], metricsSize[1], metricsSize[0]))
+#                         newMetrics[targetInt[1]-1,targetInt[0]-1] += 1
+                        
                         #TODO: make it one numpy array action:
                         for sx in range(spotSize[0]):
                             for sy in range(spotSize[1]):
@@ -539,7 +570,6 @@ while True:
      
                                 if mx >= 0 and my >= 0 and mx < metricsSize[0] and my < metricsSize[1]:
                                     newMetrics[my,mx] += spot[sx,sy] #/ 20
-                    # print("MAX",np.max(newMetrics))
     
     
                 # after we collected all new metrics, blur them foor smoothness
@@ -555,12 +585,12 @@ while True:
         if result is not None and not args.hide_preview:
             cv2.imshow("Output", im)
         te3 = time.time()
+        logger.debug("Pre processing took: {}s".format(te2-tr1))
         logger.debug("showed webcam image in %fs", te3-te2)
         logger.debug("Rendering took %fs", te3-te1)
         logger.debug("Waited took %fs", te1b-te1)
     
     
-
         # blur smooth the heatmap
         # logger.debug("Max blurred metrics: %f", np.max(metrics))
     
@@ -568,7 +598,6 @@ while True:
         tm21 = time.time()
         t = tm21
     
-
         diffT = min(1, t - lastRunTime)
         lastRunTime = t
         # animDuration = 1
@@ -581,15 +610,65 @@ while True:
         # smooth impact of first hits by having at least 0.05
         normalisedMetrics = metrics / (max(255*7 ,np.max(metrics)))
         # convert to colormap, thanks to: https://stackoverflow.com/a/10967471
+        
+        if args.only_metrics:
+            # output only metrics instead of heatmap. Usefull for debugging O:)
+            nmax = np.max(newMetrics)
+            renderMetrics = newMetrics/nmax if nmax > 0 else newMetrics
+            normalisedMetricsColored = np.uint8(renderMetrics *255 )
+            normalisedMetricsColoredBGR = cv2.cvtColor(normalisedMetricsColored, cv2.COLOR_GRAY2BGR)
+            # draw grid lines
+            for i in range(int(metricsSize[0]/100)):
+                cv2.line(normalisedMetricsColoredBGR, (i*100, 0), (i*100, metricsSize[1]), (150,150,150), 1)
+        else:
             normalisedMetricsColored = np.uint8(cm.nipy_spectral(normalisedMetrics)*255)
             normalisedMetricsColoredBGR = cv2.cvtColor(normalisedMetricsColored, cv2.COLOR_RGB2BGR)
             
+        
+        if currentPoint is not None and args.verbose:
+            cv2.putText(normalisedMetricsColoredBGR, "x: {}".format(currentPoint[0]), (10,70), cv2.FONT_HERSHEY_PLAIN, .7, (255,255,255))
+            cv2.putText(normalisedMetricsColoredBGR, "y: {}".format(currentPoint[1]), (10,90), cv2.FONT_HERSHEY_PLAIN, .7, (255,255,255))
+            
+            cv2.putText(normalisedMetricsColoredBGR, "pos: x: {}, y: {}, z: {}".format(
+                currentVectors['translation'][0],
+                currentVectors['translation'][1],
+                currentVectors['translation'][2]
+                ), (10,110), cv2.FONT_HERSHEY_PLAIN, .7, (255,255,255))
+            cv2.putText(
+                normalisedMetricsColoredBGR,
+                "rot: x: {}, y: {}, z{}".format(
+                    currentVectors['rotation'][0],
+                    currentVectors['rotation'][1],
+                    currentVectors['rotation'][2],
+                ), (10,130), cv2.FONT_HERSHEY_PLAIN, .7, (255,255,255))
+            targetPoint = transform(currentPoint)
+            logger.info("Are we really looking at {}".format(targetPoint))
+            logger.info("Size: {}".format(normalisedMetricsColoredBGR.shape))
+            cv2.circle(normalisedMetricsColoredBGR, targetPoint, 2, (0,255,0), -1)
+            cv2.line(
+                normalisedMetricsColoredBGR,
+                (metricsSize[0]/2,metricsSize[1]/2),
+                tuple(targetPoint),
+                (255,0,0), 2
+                )
+            
+#             cv2.putText(normalisedMetricsColoredBGR, "z: {}".format(currentPoint[2]), (10,70), cv2.FONT_HERSHEY_PLAIN, .7, (255,255,255))
+    
         tm22 = time.time()
         logger.debug("Max normalised metrics: %f", np.max(normalisedMetrics))
         # logger.info(normalisedMetrics)
         tm23 = time.time()
         
-    cv2.imshow("test",normalisedMetricsColoredBGR)
+#         normalisedMetricsColoredBGR = cv2.resize(normalisedMetricsColoredBGR, tuple(renderSize))
+        if backdrop is not None:
+            dx = (screenSize[0] - renderSize[0]) / 2
+            dy = (screenSize[1] - renderSize[1]) / 2
+            print(dx, dy)
+            backdrop[dy:dy+renderSize[1], dx:dx+renderSize[0]] = normalisedMetricsColoredBGR
+            renderImage = backdrop
+        else:
+            renderImage = normalisedMetricsColoredBGR
+        cv2.imshow("test", renderImage)
         # image = Image.fromarray(normalisedMetricsColored)
         # wpercent = (imageWindowSize[0] / float(image.size[0]))
         # hsize = int((float(image.size[1]) * float(wpercent)))
@@ -607,7 +686,7 @@ while True:
         # imagesprite = imageCanvas.create_image(renderSize[0]/2, renderSize[1]/2,image=tkpi, tags="IMG")
         # imageWindowRoot.update()
         tm24 = time.time()
-    logger.debug("PIL image generated in %fs", tm24 - tm23)
+        logger.debug("Render in generated in %fs", tm24 - tm23)
         # logger.debug("Total matrix time is %fs", tm4 - tm3 + tm2 - tm1 + tm24 - tm21)
     
         if not args.hide_graph:
@@ -677,10 +756,10 @@ while True:
                 pickle.dump( coordinates, open( "coordinates.p", "wb" ) )
                 logger.info("Saved coordinates")
     
-            transform = create_perspective_transform(coordinatesToSrc(coordinates), screenDrawCorners)
+                transform = create_perspective_transform(coordinatesToSrc(coordinates), screenDrawCorners, True)
     
         duration = time.time()-te1
         fps = 1/duration
         logger.info("Rendering loop %fs %ffps", duration, fps)
     
-cv2.destroyAllWindows()
+    cv2.destroyAllWindows()

output/.gitignore

@@ -1,3 +0,0 @@
-*
-!.gitignore
-