class Main:
def __init__(self):
self.stream_parent_conn, self.stream_child_conn = mp.Pipe()
model_path = '../data/faster_rcnn_inception_v2_coco_2018_01_28/frozen_inference_graph.pb'
self.odapi = DetectorAPI(path_to_ckpt=model_path)
self.cal = Calibration()
self.comm = Comm()
self.stream = Stream(self.stream_child_conn)
self.str = mp.Process(target=self.stream.run, args=())
self.eat = None
self.socket_update = None
self.frame_lock = mp.RLock()
self.coords_lock = mp.RLock()
self.coords = []
def totuple(self, a):
try:
return tuple(self.totuple(i) for i in a)
except TypeError:
return a
def start(self):
self.str.start()
self.eat = th.Thread(target=self.eat_rest, name="eat")
self.socket_update = th.Thread(target=self.update_game, name="update")
self.eat.start()
self.socket_update.start()
ref = cv2.imread('../pics/soccerfield_2d.png')
height, width, cols = ref.shape
self.h = None
pts_src = None
refClean = ref.copy()
self.frame_lock.acquire()
img = self.stream_parent_conn.recv()
self.frame_lock.release()
try:
pts_src = np.load("../data/calibrated_3D.npy")
pts_dst = np.float32([[0, 0], [width, 0], [0, height], [width, height]])
self.h, status = cv2.findHomography(pts_src, pts_dst)
except Exception as e:
print(e)
self.coords_3d = self.cal.manual_calibrate(img)
if len(self.coords_3d) is 4:
pts_dst = np.float32([[0, 0], [width, 0], [0, height], [width, height]])
pts_src = np.float32(self.coords_3d)
self.h, status = cv2.findHomography(pts_src, pts_dst)
np.save("../data/calibrated_3D", pts_src)
coords = []
while True:
if not self.str.is_alive():
print("str died")
break
start_time = time.time()
self.frame_lock.acquire()
img = self.stream_parent_conn.recv()
self.frame_lock.release()
x, y, w, h = cv2.boundingRect(pts_src)
img = img[y-CropOffset:y + h, x:x + w].copy()
boxes, scores, classes, num = self.odapi.processFrame(img)
# Visualization of the results of a detection.
ref = refClean.copy()
for i in range(len(boxes)):
# Class 1 represents human
if classes[i] == 1 and scores[i] > threshold:
if len(coords) > 8:
break
box = boxes[i]
cv2.rectangle(img, (box[1], box[0]), (box[3], box[2]), (255, 0, 0), 2)
w = box[3] - box[1]
h = box[2] - box[0]
middle_point = int(box[1]+(w/2)), int(box[2]-h/2)
cv2.circle(img, (middle_point), 4, (0, 0, 255), 2)
a = np.array([[middle_point[0]+x, middle_point[1]+y-CropOffset+25]], dtype='float32') # 25 pixels naar beneden om te compenseren voor de halve lengte persoon
a = np.array([a])
relative_coords = cv2.perspectiveTransform(a, self.h)
rx, ry = self.totuple(relative_coords[0][0])
cv2.circle(ref, (rx, ry), 10, (0, 0, 255), 4)
padding = 1.15
if rx > (width-(width*padding)) and rx < (width*padding) and ry > (height-(height*padding)) and ry < (height*padding):
coords.append(((rx/width) * 16, 9-((ry/height)*9)))
#cv2.imshow("transform", ref)
cv2.imshow("preview", img)
for x in range(len(coords), 8, 1):
coords.append((0,0))
self.coords_lock.acquire()
self.coords = coords
self.coords_lock.release()
coords = []
end_time = time.time()
#print("Elapsed Time:", end_time - start_time)
if cv2.waitKey(1) & 0xFF == ord('q'):
self.str.terminate()
self.str.join()
break
self.eat.join()
self.socket_update.join()
return
def eat_rest(self):
print("Eating started...")
while True:
self.frame_lock.acquire()
x = self.stream_parent_conn.recv()
self.frame_lock.release()
if not self.str.is_alive():
break
def update_game(self):
print("Update over socket started...")
com = Comm()
coords = []
while True:
if com.open():
while True:
self.coords_lock.acquire()
coords = self.coords
self.coords_lock.release()
if len(coords) > 8:
coords = coords[0:8]
elif len(coords) < 8:
for x in range(len(coords), 8, 1):
coords.append((0, 0))
com.send(coords)
time.sleep(0.1)
if not self.str.is_alive():
break
com.close()
if not self.str.is_alive():
break
com.close()
