class actions(object):
def __init__(self, arguments):
self.arguments = arguments
# Frame window dim
self.winWidth = 640
self.winHeight = 480
actionPredictor_params.__init__(self)
self.fps_time = 0
#self.step = 15
self.mode = {
'Pose Estimation': 'estimation',
'Tracking': 'tracking',
'Action Recognition': 'recognition'
}
w, h = model_wh(self.arguments.resize)
if w > 0 and h > 0:
self.estimator = TfPoseEstimator(get_graph_path(
self.arguments.model),
target_size=(w, h))
else:
self.estimator = TfPoseEstimator(get_graph_path(
self.arguments.model),
target_size=(432, 368))
self.cam = cv2.VideoCapture(self.arguments.camera)
# Tracker based on Sort
self.sort_max_age = 20
self.sort_min_hit = 3
self.tracker = Sort(self.sort_max_age, self.sort_min_hit)
def proceed(self):
self._read_frame_()
if self.ret_val and self.arguments.mode == self.mode['Pose Estimation']:
self._perform_estimation_()
elif self.ret_val and self.arguments.mode == self.mode['Tracking']:
self._perform_tracking_()
elif self.ret_val and self.arguments.mode == self.mode[
'Action Recognition']:
self._perform_recognition_()
else:
sys.exit(
'Abort...please choose correct action mode from "estimation" "tracking" "recognition"'
)
self._output_()
def _joint_filter_(self, joint):
if 1 not in joint:
return False
# Check exist of hip
if 8 not in joint and 11 not in joint:
return False
# Check exist of knee
if 9 not in joint and 12 not in joint:
return False
return True
def _joint_complete_(self, joint):
if 8 in joint and 11 not in joint:
joint[11] = joint[8]
elif 8 not in joint and 11 in joint:
joint[8] = joint[11]
if 9 in joint and 12 not in joint:
joint[12] = joint[9]
elif 9 not in joint and 12 in joint:
joint[9] = joint[12]
return joint
def _perform_estimation_(self):
self.humans = self.estimator.inference(self.image)
self.image = TfPoseEstimator.draw_humans(self.image,
self.humans,
imgcopy=False)
def _perform_tracking_(self):
self.humans = self.estimator.inference(self.image)
self.image, joints, bboxes, xcenter, sk = TfPoseEstimator.get_skeleton(
self.image, self.humans, imgcopy=False)
height = self.image.shape[0]
width = self.image.shape[1]
if bboxes:
result = np.array(bboxes)
det = result[:, 0:5]
det[:, 0] = det[:, 0] * width
det[:, 1] = det[:, 1] * height
det[:, 2] = det[:, 2] * width
det[:, 3] = det[:, 3] * height
trackers = self.tracker.update(det)
for d in trackers:
xmin = int(d[0])
ymin = int(d[1])
xmax = int(d[2])
ymax = int(d[3])
label = int(d[4])
cv2.rectangle(
self.image, (xmin, ymin), (xmax, ymax),
(int(self.c[label % 32, 0]), int(
self.c[label % 32, 1]), int(self.c[label % 32, 2])), 4)
def _perform_recognition_(self):
self.predictor = actionPredictor()
self.humans = self.estimator.inference(self.image)
self.image, joints, bboxes, xcenter, sk = TfPoseEstimator.get_skeleton(
self.image, self.humans, imgcopy=False)
height = self.image.shape[0]
width = self.image.shape[1]
if bboxes:
result = np.array(bboxes)
det = result[:, 0:5]
det[:, 0] = det[:, 0] * width
det[:, 1] = det[:, 1] * height
det[:, 2] = det[:, 2] * width
det[:, 3] = det[:, 3] * height
trackers = self.tracker.update(det)
self.current = [i[-1] for i in trackers]
if len(self.previous) > 0:
for item in self.previous:
if item not in self.current and item in self.data:
del self.data[item]
if item not in self.current and item in self.memory:
del self.memory[item]
self.previous = self.current
for d in trackers:
xmin = int(d[0])
ymin = int(d[1])
xmax = int(d[2])
ymax = int(d[3])
label = int(d[4])
#logger.debug('label is: %d' % (label))
# Locate the current person object in current frame
# Iterated thru xcenter for finding minimum distance between object center coord
try:
j = np.argmin(
np.array(
[abs(i - (xmax + xmin) / 2.) for i in xcenter]))
except:
j = 0
# Check if major skeleton points are existing
if self._joint_filter_(joints[j]):
joints[j] = self._joint_complete_(
self._joint_complete_(joints[j]))
if label not in self.data:
#logger.debug('label is: %d' % (label))
self.data[label] = [joints[j]]
self.memory[label] = 0
else:
self.data[label].append(joints[j])
if len(self.data[label]) == self.step:
pred = self.predictor.move_status(self.data[label])
#logger.debug(len(self.data[label]))
if pred == 0:
pred = self.memory[label]
else:
self.memory[label] = pred
self.data[label].pop(0)
location = self.data[label][-1][1]
#location = functools.reduce(lambda x, y: x + y, self.data[label][:][1]) / len(self.data[label][:][1])
#location = sum(self.data[label][:][1]) / float(len(self.data[label][:][1]))
if location[0] <= 30:
location = (51, location[1])
if location[1] <= 10:
location = (location[0], 31)
cv2.putText(self.image, self.move_status[pred],
(location[0] - 30, location[1] - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.8,
(255, 255, 255), 2)
cv2.rectangle(
self.image, (xmin, ymin), (xmax, ymax),
(int(self.c[label % 32, 0]), int(
self.c[label % 32, 1]), int(self.c[label % 32, 2])), 4)
def _read_frame_(self):
self.ret_val, self.image = self.cam.read()
self.image = cv2.resize(self.image, (self.winWidth, self.winHeight))
def _output_(self):
# Calculate frame averaging step
FPS = float(1.0 / (time.time() - self.fps_time))
#logger.debug('FPS: %f' % FPS)
#self.step = int(0.7 * FPS)
#logger.debug('step: %d' % self.step)
cv2.putText(self.image,
"FPS: %f" % (1.0 / (time.time() - self.fps_time)),
(10, 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255),
2)
cv2.imshow('tf-pose-estimation result', self.image)
self.fps_time = time.time()
import settings
from pose.estimator import TfPoseEstimator
from pose.networks import get_graph_path
from imutils.video import VideoStream
poseEstimator = None
poseEstimator = TfPoseEstimator(get_graph_path('mobilenet_thin'), target_size=(432, 368))
cap=cv2.VideoCapture(0)
#cap = VideoStream(src='rtsp://*****:*****@192.168.51.162/PSIA/streaming/channels/102').start()
while True:
ret,frame=cap.read()
ret=True
if ret :
show = cv2.resize(frame, (settings.winWidth, settings.winHeight))
humans = poseEstimator.inference(show)
show = TfPoseEstimator.draw_humans(show, humans, imgcopy=False)
cv2.imshow("frame",show)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
if not os.path.exists(root_path + action + '/txt/'):
os.mkdir(root_path + action + '/txt/')
os.mkdir(root_path + action + '/imgs/')
samples = len(os.listdir(root_path + action + '/txt/'))
sample_count = 1000 if samples == 0 else 1000 + samples
e = TfPoseEstimator(get_graph_path('mobilenet_thin'),
target_size=(432, 368))
cam = cv2.VideoCapture(0)
ret_val, image = cam.read()
joints = []
joints_imgs = []
while True:
ret_val, image = cam.read()
if ret_val:
humans = e.inference(image)
image, joint, *_, sk = TfPoseEstimator.get_humans(image,
humans,
imgcopy=False)
if joint:
if len(joints) < clip_length:
joints.append(joint[0])
joints_imgs.append(sk)
else:
joints.pop(0)
joints_imgs.pop(0)
joints.append(joint[0])
joints_imgs.append(sk)
cv2.putText(image, "FPS: %.2f" % (1.0 / (time.time() - fps_time)),
(10, 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0),
2)