def callback_image(data):
# print 'lam on day viet anh'
global skip, i_left, i_right, image, color_image, rects
img = convert_data_to_image(data.data)
if skip > 0:
return
cv2.line(img, (i_left, 0), (i_left, 159), (0, 255, 0), 2)
cv2.line(img, (i_right, 0), (i_right, 159), (255, 0, 0), 2)
# cv2.line(img, (0, int(6 * 160 / 16)), (0, int(6 * 160 / 16)), (255, 0, 0), 2)
# cv2.line(img, (0, int(12 * 160 / 16)), (319, int(12 * 160 / 16)), (255, 0, 0), 2)
color_image = img
img_cpy = np.expand_dims(img / 255., axis=0)
seg1 = model1.predict(img_cpy)[0]
seg1 = np.argmax(seg1, axis=2)
image = seg1
color = (0, 0, 255)
# if len(rects) > 0:
# for rect in rects:
# x, y, w, h = rect
#
# if obstackle_classify(image * 1., rect):
# cv2.rectangle(img, (x, y), (x + w, y + h), (0, 255, 0), 1)
# else:
# cv2.rectangle(img, (x, y), (x + w, y + h), color, 1)
seg1 = fill(np.uint8(seg1))
seg1 = seg1 * 255.
image = seg1
# _, _, res = s2s.get_steer(img, seg1, 0, 0, i_left, i_right)
# cv2.imshow('image', img)
# cv2.imshow('seg', res)
cv2.waitKey(1)
def callback(self, data):
global skip
if skip > 0:
skip -= 1
return
global is_running
# global check
# global start
# global t1
# if check == True:
# start = time.time()
# check = False
try:
self.image = self.convert_data_to_image(data.data)
# if self.frame % 6 == 0:
# self.flag, s = self.sign_model.predict(self.image)
# self.frame = 0
# print self.flag
self.image = cv2.resize(self.image, (320, 160))
res, sharp = self.model.predict(self.image)
res = fill(np.uint8(res))
self.pred_image = res * 1.
# print self.left_restriction, self.right_restriction
speed, steer = self.s2s.get_steer(res, self.flag, sharp,
self.left_restriction,
self.right_restriction)
if self.s2s.park_time and not self.s2s.on_crossroad:
parking_speed = self.parking(self.image)
if parking_speed > -1:
self.s2s.speed_memory.pop()
self.s2s.speed_memory.append(parking_speed)
speed = self.s2s.mean_speed_queue()
speed -= 1
# cv2.imshow('black and white', self.image)
# cv2.waitKey(1)
cv2.imshow('road', res * 255.)
cv2.waitKey(1)
# print (1 / (time.time() - t1))
if is_running:
self.publish_data(speed, -steer)
else:
# self.s2s.total_width = self.s2s.roi * 160
self.s2s.total_time_steer = 0.0
self.total_time = 0.0
self.s2s.mode = 0
self.s2s.counter = 0
self.s2s.reset_actions()
self.s2s.speed_memory = deque(iterable=np.zeros(
5, dtype=np.uint8),
maxlen=5)
self.s2s.error_proportional_ = 0.0
self.s2s.error_integral_ = 0.0
self.s2s.error_derivative_ = 0.0
self.publish_data(0, 0)
# print 1/(time.time() - t1)
# t1 = time.time()
# self.frame += 1
except CvBridgeError as e:
print(e)
def get_points(self, img):
segmented_img, _ = self.model.predict(img)
floodfilled_img = fill(np.uint8(segmented_img))
#cv2.imshow('segmented image', floodfilled_img * 255.)
#cv2.waitKey(1)
floodfilled_img = floodfilled_img.astype(np.int32)
x_left, x_right = p2c_main.get_center_points_by_roi(
floodfilled_img, self.roi)
return x_left, x_right
def getDistance1(heatmap,groundtruth,thre):
distance = np.zeros((1,RANGE))
tmp = getGazeHeatmap(heatmap, thre)
dis_t = []
region, ht = floodfill.fill(tmp)
pred_center = []
gt_center = []
if len(region) != 0:
for iter in range(len(region)):
a = np.asarray(region[iter])
x = sum(a[:, 0]) / float(len(a[:, 0]))
y = sum(a[:, 1]) / float(len(a[:, 1]))
pred_center.append([x, y])
pred = np.asarray(pred_center)
gt, gt_ht = floodfill.fill(groundtruth)
if len(gt) != 0:
for iter in range(len(gt)):
a = np.asarray(gt[iter])
x = sum(a[:, 0]) / float(len(a[:, 0]))
y = sum(a[:, 1]) / float(len(a[:, 1]))
gt_center.append([x, y])
if len(gt_center) == 0:
return distance
else:
while gt_center:
if len(pred) == 0:
for l in range(len(gt_center)):
dis_t.append(420) #if there's no potential predicted point,return 420
break
a = gt_center.pop()
# for jter in range(len(gt_center)):
# a = np.asarray(gt_center[jter])
b = pred - a
diff = np.sqrt(np.square(b[:,0]) + np.square(b[:,1]))
pos = diff.argmin()
dis_t.append(diff.min())
pred = np.delete(pred, (0), axis=0)
return float(sum(dis_t))
def getDistance(heatmap, groundtruth):
distance = np.zeros((1, RANGE))
for i in range(RANGE):
thre = float(1) / float(RANGE) * float(i + 1)
tmp = getGazeHeatmap(heatmap, thre)
dis_t = []
region, ht = floodfill.fill(tmp)
pred_center = []
gt_center = []
if len(region) != 0:
for iter in range(len(region)):
a = np.asarray(region[iter])
x = sum(a[:, 0]) / float(len(a[:, 0]))
y = sum(a[:, 1]) / float(len(a[:, 1]))
pred_center.append([x, y])
pred = np.asarray(pred_center)
gt, gt_ht = floodfill.fill(groundtruth)
if len(gt) != 0:
for iter in range(len(gt)):
a = np.asarray(gt[iter])
x = sum(a[:, 0]) / float(len(a[:, 0]))
y = sum(a[:, 1]) / float(len(a[:, 1]))
gt_center.append([x, y])
if len(gt_center) == 0:
return distance
else:
while gt_center:
if len(pred) == 0:
for l in range(len(gt_center)):
dis_t.append(420) # if there's no potential predicted point,return 420
break
a = gt_center.pop()
# for jter in range(len(gt_center)):
# a = np.asarray(gt_center[jter])
b = pred - a
diff = np.sqrt(np.square(b[:, 0]) + np.square(b[:, 1]))
pos = diff.argmin()
dis_t.append(diff.min())
pred = np.delete(pred, (0), axis=0)
distance[0, i] = float(sum(dis_t))
return distance
# plt.figure()
#
# plt.subplot(1, 2, 1)
# plt.title('original heatmap')
# plt.imshow(tmp)
#
# plt.subplot(1, 2, 2)
# plt.title('clustered heatmap')
# plt.imshow(ht)
# plt.axis('off')
#
# plt.show()
# heatmap = np.loadtxt('hm.txt')
# thre = 1.2
# groundtruth = tmp = getGazeHeatmap(heatmap, 1.7)
# dis = getDistance(heatmap, thre, groundtruth)
# print(dis)