def main():
relation_vector = np.array(json.loads(args.relation_zones))
contrast_vector = np.array(json.loads(args.contrast))
gamma_vector = np.array(json.loads(args.gamma))
arguments_com(args)
model = posenet.load_model(args.model)
model = model.cuda()
output_stride = model.output_stride
#model_cfg, model_outputs = posenet.load_model(args.model, sess)
#output_stride = model_cfg['output_stride']
if args.cam_or_file == 0:
cap = cv2.VideoCapture(args.input_file_name)
elif args.cam_or_file == 1:
cap = cv2.VideoCapture(args.cam_id)
fourcc = cv2.VideoWriter_fourcc(*'DIVX')
height = cap.get(4)
width = cap.get(3)
listofcenters = []
centers = []
max_displacement = []
max_movement_single = []
faceTrackers = {}
frame_number = 0
currentFaceID = 0
cont = 0
width_min = 0
min_fid = 0
zones_computer = [0, 0, 0]
width_int = width / 3
rectangleColor = (0, 0, 255)
height_fin = height * (8 / 10)
height_min = height * (2 / 7)
h_up, h_down, w_up, w_down, number_partition, zones = division_funct(
height, width, args)
out = cv2.VideoWriter(
args.output_file_name, fourcc, 20.0,
(int(width), int(math.ceil(height_fin - height_min))))
len_video = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
start = time.time()
zone_number = []
cnt_6 = 0
frame_count = 0
while True:
if (int(time.time() - start) > (args.prueba)) and (args.prueba != 0):
exit()
if int(time.time() - start) > ((cont + 1) * args.seconds_movement):
zone_number = wifi_data(args)
if args.print_data == 1:
print('Average FPS: ', frame_count / (time.time() - start))
print('time: ', (time.time() - start))
cont = print_move(cont, max_displacement, max_movement_single,
min_fid)
if frame_count == len_video:
break
res, img = cap.read()
print(img.shape)
cnt = 0
my_batch = list()
cnt = cnt + 1
res, img = cap.read()
frame_count += 1
if not res:
break
img = preprocessing(h_down, h_up, w_down, w_up, gamma_vector, img,
contrast_vector, args, height_min, height_fin,
width)
print(img.shape)
input_image, display_image, output_scale = posenet.process_input(
img, scale_factor=args.scale_factor, output_stride=output_stride)
print(input_image.shape)
if not res:
break
cnt = 0
fidsToDelete = []
number_partition = zeros_number_partition(args)
if args.print_data == 1:
print("Number of people detected: ", len(faceTrackers))
for fid in faceTrackers.keys():
cnt += 1
trackingQuality = faceTrackers[fid].update(overlay_image)
if trackingQuality < args.track_quality:
if args.reset_movement == 1:
del max_movement_single[cnt]
del max_displacement[cnt]
fidsToDelete.append(fid)
for fid in fidsToDelete:
faceTrackers.pop(fid, None)
# test_2 = list([input_image, input_image])
# test_3 = np.stack(test_2, axis=0)
# print(input_image.shape,img.shape)
input_image = torch.Tensor(input_image).cuda()
heatmaps_result, offsets_result, displacement_fwd_result, displacement_bwd_result = model(
input_image)
pose_scores, keypoint_scores, keypoint_coords = posenet.decode_multi.decode_multiple_poses(
heatmaps_result.squeeze(0),
offsets_result.squeeze(0),
displacement_fwd_result.squeeze(0),
displacement_bwd_result.squeeze(0),
output_stride=output_stride,
max_pose_detections=10,
min_pose_score=0.15)
keypoint_coords *= output_scale
if args.part_shown == 0:
overlay_image = posenet.draw_skel_and_kp(display_image,
pose_scores,
keypoint_scores,
keypoint_coords,
min_pose_score=0.15,
min_part_score=0.1)
elif args.part_shown == 1:
overlay_image = face_calculation(keypoint_coords,
keypoint_scores,
display_image,
pose_scores,
min_pose_score=0.15,
min_part_score=0.1)
elif args.part_shown == 2:
overlay_image = posenet.draw_keypoints(display_image,
pose_scores,
keypoint_scores,
keypoint_coords,
min_pose_score=0.15,
min_part_score=0.15)
elif args.part_shown == 3:
overlay_image = posenet.draw_skeleton(display_image,
pose_scores,
keypoint_scores,
keypoint_coords,
min_pose_score=0.15,
min_part_score=0.15)
elif args.part_shown == 4:
overlay_image = arms_calculation(keypoint_coords,
keypoint_scores,
display_image,
pose_scores,
min_pose_score=0.15,
min_part_score=0.1)
elif args.part_shown == 5:
overlay_image = mean_calculation(keypoint_coords,
keypoint_scores,
display_image,
pose_scores,
min_pose_score=0.15,
min_part_score=0.1)
elif args.part_shown == 6:
overlay_image = chest_calculation(keypoint_coords,
keypoint_scores,
display_image,
pose_scores,
min_pose_score=0.15,
min_part_score=0.1)
elif args.part_shown == 7:
overlay_image = att_calculation(keypoint_coords,
keypoint_scores,
display_image,
pose_scores,
min_pose_score=0.15,
min_part_score=0.1)
if (frame_count % args.n_frames) == 0:
for idx, esq in enumerate(keypoint_coords):
x_coords = []
y_coords = []
for ii in range(0, min(args.detection_zone, len(esq) - 1)):
if esq[ii][0] != 0 or esq[ii][1] != 0:
x_coords.append(esq[ii][0])
y_coords.append(esq[ii][1])
if len(x_coords) != 0:
x_min = np.min(x_coords) - 20
y_min = np.min(y_coords) - 20
x_max = np.max(x_coords) + 20
y_max = np.max(y_coords) + 20
x = x_min
y = y_min
w = x_max - x_min
h = y_max - y_min
x_bar = x + 0.5 * w
y_bar = y + 0.5 * h
matchedFid = None
for fid in faceTrackers.keys():
tracked_position = faceTrackers[fid].get_position()
t_x = int(tracked_position.left())
t_y = int(tracked_position.top())
t_w = int(tracked_position.width())
t_h = int(tracked_position.height())
t_x_bar = t_x + 0.5 * t_w
t_y_bar = t_y + 0.5 * t_h
if ((t_y <= x_bar <= (t_y + t_h)) and (t_x <= y_bar <=
(t_x + t_w))
and (x <= t_y_bar <=
(x + w)) and (y <= t_x_bar <= (y + h))):
matchedFid = fid
centers = listofcenters[fid]
centers = [(int(y_bar), int(x_bar))] + centers
listofcenters[fid] = centers
max_distance = max_displacement[fid]
for (x, y) in centers:
distance = abs(
(pow(x_bar, 2) + pow(y_bar, 2)) -
(pow(t_x_bar, 2) + pow(t_y_bar, 2)))
max_displacement[fid] += distance
if distance > max_movement_single[fid]:
max_movement_single[fid] = distance
if ((matchedFid is None)):
tracker = dlib.correlation_tracker()
tracker.start_track(
overlay_image,
dlib.rectangle(int(y_min), int(x_min), int(y_max),
int(x_max)))
faceTrackers[currentFaceID] = tracker
currentFaceID += 1
centers = []
centers = [(int(y_bar), int(x_bar))] + centers
listofcenters.append(centers)
max_displacement.append(0)
max_movement_single.append(0)
cnt_6 += 1
if cnt_6 == args.new_data:
for fid in faceTrackers.keys():
fidsToDelete.append(fid)
for fid in fidsToDelete:
faceTrackers.pop(fid, None)
min_fid = fid
cnt_6 = 0
for fid in faceTrackers.keys():
tracked_position = faceTrackers[fid].get_position()
t_x = int(tracked_position.left())
t_y = int(tracked_position.top())
t_w = int(tracked_position.width())
t_h = int(tracked_position.height())
t_x_bar = int(t_x + 0.5 * t_w)
t_y_bar = int(t_y + 0.5 * t_h)
centers = listofcenters[fid]
centers = [(t_x_bar, t_y_bar)] + centers
listofcenters[fid] = centers
if args.reset_movement == 1:
max_distance = max_displacement[fid - min_fid]
else:
max_distance = max_displacement[fid]
for (x, y) in centers:
distance = abs((pow(x_bar, 2) + pow(y_bar, 2)) -
(pow(t_x_bar, 2) + pow(t_y_bar, 2)))
if args.reset_movement == 1:
max_displacement[fid - min_fid] += distance
if distance > max_movement_single[fid - min_fid]:
max_movement_single[fid - min_fid] = distance
else:
max_displacement[fid] += distance
if distance > max_movement_single[fid]:
max_movement_single[fid] = distance
if args.show_image == 1 or args.store_image == 1:
cv2.rectangle(overlay_image, (int(t_x), int(t_y)),
(int(t_x + t_w), int(t_y + t_h)), (0, 255, 0), 2)
cv2.circle(overlay_image, (int(t_x_bar), int(t_y_bar)), 5,
(0, 255, 0), 2)
zones_computer, number_partition = zone_calculation(
zones, t_x_bar, t_y_bar, number_partition, height_min,
relation_vector)
if args.print_data == 1:
print("The number of people in each partition is: ",
number_partition)
print(
"The number of people in each zon from the computer vision is :",
zones_computer)
if zone_number != []:
for ll in range(0, len(zones_computer)):
if (zones_computer[ll] <
(args.threshold_zones * zone_number[ll])):
#contrast_vector[ll] +=1
#gamma_vector[ll] +=0.25
a = 1
show_store_result(args, overlay_image, faceTrackers, listofcenters)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
print('Average FPS: ', frame_count / (time.time() - start))
print('time: ', (time.time() - start))
cont = print_move(cont, max_displacement, max_movement_single, min_fid)
def main():
contrast_vector=np.array(json.loads(args.contrast))
gamma_vector=np.array(json.loads(args.gamma))
if args.grid_size !=4 and args.grid_size !=2 and args.grid_size !=1:
print("The grid size must be either 1,2 or 4 for better results")
exit()
elif args.detection_zone!=5 and args.detection_zone!=11 and args.detection_zone!=17:
print("The detection zone must be either 5 (for face), 11 (for upper body) or 17 (for whole body)")
exit()
elif args.track_quality!=7 and args.track_quality!=8:
print("The tracking quality must be either 7 or 8 for better results")
exit()
elif np.mean(contrast_vector)<5 or np.mean(contrast_vector)>15:
print("The contrast must be between 5 and 15 for better results")
exit()
elif np.mean(gamma_vector)<2.75 or np.mean(gamma_vector)>5.0:
print("The gamma correction factor must be between 2.75 and 5.0 for better results")
exit()
elif args.cam_or_file !=0 and args.cam_or_file!=1:
print("Select the source of the video 0 for file and 1 for camera")
exit()
elif args.cam_or_file ==0 and args.input_file_name=="a":
print("Introduce a file name")
exit()
elif args.n_divisions != 6 and args.n_divisions !=9:
print("number of division must be either 6 or 9",args.n_divisions)
exit()
elif args.n_divisions !=len(contrast_vector) or args.n_divisions!=len(gamma_vector):
print("Number of divisions must be equal to length of gamma and contrast vectors")
exit()
with tf.Session() as sess:
model_cfg, model_outputs = posenet.load_model(args.model, sess)
output_stride = model_cfg['output_stride']
if args.cam_or_file == 0:
cap = cv2.VideoCapture(args.input_file_name)
elif args.cam_or_file==1:
cap = cv2.VideoCapture(args.cam_id)
fourcc = cv2.VideoWriter_fourcc(*'DIVX')
height = cap.get(4)
width = cap.get(3)
listofcenters = []
centers = []
max_displacement = []
max_movement_single = []
frame_number = 0
currentFaceID = 0
cont = 0
rectangleColor = (0,0,255)
faceTrackers = {}
height_fin = height*(8/10)
width_int = width/3
height_min = height*(2/7)
width_min = 0
if args.n_divisions==6:
height_int = (height_min + height_fin)/2
h_up = [height_int,height_int,height_int,height_fin,height_fin,height_fin]
h_down = [height_min,height_min,height_min,height_int,height_int,height_int]
elif args.n_divisions==9:
height_int = (height_fin - height_min)/3
h_up = [height_min + height_int, height_min + height_int, height_min + height_int, height_min + 2*height_int, height_min + 2*height_int, height_min + 2*height_int, height_fin, height_fin, height_fin]
h_down = [height_min, height_min, height_min, height_min + height_int, height_min + height_int, height_min + height_int, height_min + 2*height_int, height_min + 2*height_int, height_min + 2*height_int]
w_up = [width_int,width_int*2,(width_int*3),width_int,(width_int*2),(width_int*3),width_int,(width_int*2),(width_int*3)]
w_down = [width_min,width_int,width_int*2,width_min,width_int,width_int*2,width_min,width_int,width_int*2]
out = cv2.VideoWriter(args.output_file_name, fourcc, 20.0, (int(width),int(math.ceil(height_fin-height_min))))
len_video = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
start = time.time()
frame_count = 0
print("Nueva")
while True:
if (int(time.time()-start)> (args.prueba)) and (args.prueba!=0):
exit()
if int(time.time()-start)> ((cont+1)*args.seconds_movement):
print('Average FPS: ', frame_count / (time.time() - start))
print('time: ', (time.time() - start))
cont=print_move(cont,max_displacement,max_movement_single)
if frame_count==len_video:
break
res, img = cap.read()
if not res:
break
for ii in range(0,len(h_up)):
img_1 = img[int(h_down[ii]):int(h_up[ii]),int(w_down[ii]):int(w_up[ii])]
img_1 = adjust_gamma(img_1, gamma=gamma_vector[ii]) # input
lab = cv2.cvtColor(img_1, cv2.COLOR_BGR2LAB)
l, a, b = cv2.split(lab)
clahe = cv2.createCLAHE(clipLimit=contrast_vector[ii], tileGridSize=(args.grid_size,args.grid_size))
cl = clahe.apply(l)
limg = cv2.merge((cl,a,b))
img[int(h_down[ii]):int(h_up[ii]),int(w_down[ii]):int(w_up[ii])] = cv2.cvtColor(limg, cv2.COLOR_LAB2BGR)
img = img[int(height_min):int(height_fin),0:int(width)]
input_image, display_image, output_scale = posenet.process_input(img, scale_factor=args.scale_factor, output_stride=output_stride)
frame_count += 1
fidsToDelete = []
print("Number of people detected: ",len(faceTrackers))
for fid in faceTrackers.keys():
trackingQuality = faceTrackers[fid].update(overlay_image)
if trackingQuality < args.track_quality:
fidsToDelete.append(fid)
for fid in fidsToDelete:
#print("Removing fid " + str(fid) + " from list of trackers")
faceTrackers.pop(fid,None)
if (frame_number % args.n_frames) == 0:
heatmaps_result, offsets_result, displacement_fwd_result, displacement_bwd_result = sess.run(model_outputs,feed_dict={'image:0': input_image})
pose_scores, keypoint_scores, keypoint_coords = posenet.decode_multi.decode_multiple_poses( heatmaps_result.squeeze(axis=0), offsets_result.squeeze(axis=0), displacement_fwd_result.squeeze(axis=0), displacement_bwd_result.squeeze(axis=0), output_stride=output_stride, max_pose_detections=10, min_pose_score=0.15)
keypoint_coords *= output_scale
Keypoints_face_coords = []
Keypoints_face_scores = []
Keypoints_right_arm_coords = []
Keypoints_right_arm_scores = []
Keypoints_left_arm_coords = []
Keypoints_left_arm_scores = []
Keypoints_chest_coords = []
Keypoints_chest_scores = []
Keypoints_eyes_coords = []
Keypoints_eyes_scores = []
Keypoints_half_eyes_coords = []
Keypoints_half_eyes_scores = []
Keypoints_angle_coords = []
Keypoints_angle_scores = []
Keypoints_direction = []
Keypoints_att_show_coords = []
Keypoints_att_show_scores = []
a_right_arm = []
b_right_arm = []
a_left_arm = []
b_left_arm = []
a_chest = []
b_chest = []
a_att = []
b_att = []
mean_points_face_coords = []
mean_points_face_scores = []
mean_points_chest_coords = []
mean_points_chest_scores = []
mean_points_right_arm_coords = []
mean_points_right_arm_scores = []
mean_points_left_arm_coords = []
mean_points_left_arm_scores = []
total_means_coords = []
total_means_scores = []
means_prov_coords = []
means_prov_scores = []
angle_2 = 0
angle = 0
for d in keypoint_coords:
Keypoints_face_coords.append(d[0:5])
Keypoints_eyes_coords.append(d[1:3])
for d in keypoint_scores:
Keypoints_face_scores.append(d[0:5])
Keypoints_eyes_scores.append(d[1:3])
for d in Keypoints_eyes_coords:
Keypoints_half_eyes_coords.append(np.mean(d,axis=0))
angle = math.atan2(d[1][1]-d[0][1], d[1][0]-d[0][0])
Keypoints_angle_coords.append(angle)
for d in Keypoints_eyes_scores:
Keypoints_half_eyes_scores.append(np.mean(d,axis=0))
Keypoints_angle_scores.append(np.mean(d,axis=0))
for ii in range(0,len(keypoint_coords)-1):
if Keypoints_eyes_coords[ii][0][0] == 0 and Keypoints_eyes_coords[ii][0][1] == 0 and Keypoints_eyes_coords[ii][1][0] == 0 and Keypoints_eyes_coords[ii][1][1] == 0:
Keypoints_direction.append("no hay ojos") #rectoº
continue
elif Keypoints_eyes_coords[ii][0][0] == 0 and Keypoints_eyes_coords[ii][0][1] == 0:
Keypoints_direction.append("ojo izquierdo (Derecha)") #rectoº
continue
elif Keypoints_eyes_coords[ii][1][0] == 0 and Keypoints_eyes_coords[ii][1][1] == 0:
Keypoints_direction.append("ojo derecho (Izquierda)") #rectoº
continue
if Keypoints_face_coords[ii][0][1] == 0 and Keypoints_face_coords[ii][0][0] == 0 and Keypoints_half_eyes_coords[ii][0] == 0 and Keypoints_half_eyes_coords[ii][1] == 0:
Keypoints_direction.append("fallo") #rectoº
continue
angle_2=math.atan2(Keypoints_face_coords[ii][0][1]-Keypoints_half_eyes_coords[ii][1], Keypoints_face_coords[ii][0][0]-Keypoints_half_eyes_coords[ii][0])
if abs(angle_2) < 0.261799:
Keypoints_direction.append("recto") #rectoº
elif angle_2 > 0.261799:
Keypoints_direction.append("derecha") #derecha
elif angle_2 < -0.261799:
Keypoints_direction.append("izquierda") #izq
for d in range(0,len(keypoint_coords)):
a_att.append(Keypoints_face_coords[ii][1])
a_att.append(Keypoints_face_coords[ii][2])
a_att.append(Keypoints_face_coords[ii][0])
a_att.append(Keypoints_half_eyes_coords[ii])
Keypoints_att_show_coords.append(np.asarray(a_att))
for ii in range(0,len(keypoint_scores)):
b_att.append(Keypoints_face_scores[ii][1])
b_att.append(Keypoints_face_scores[ii][2])
b_att.append(Keypoints_face_scores[ii][0])
b_att.append(Keypoints_half_eyes_scores[ii])
Keypoints_att_show_scores.append(np.asarray(b_att))
for d in keypoint_coords:
a_right_arm.append(d[6])
a_right_arm.append(d[8])
a_right_arm.append(d[10])
a_left_arm.append(d[5])
a_left_arm.append(d[7])
a_left_arm.append(d[9])
a_chest.append(d[5])
a_chest.append(d[6])
a_chest.append(d[11])
a_chest.append(d[12])
Keypoints_chest_coords.append(np.asarray(a_chest))
Keypoints_left_arm_coords.append(np.asarray(a_left_arm))
Keypoints_right_arm_coords.append(np.asarray(a_right_arm))
for d in keypoint_scores:
b_right_arm.append(d[6])
b_right_arm.append(d[8])
b_right_arm.append(d[10])
b_left_arm.append(d[5])
b_left_arm.append(d[7])
b_left_arm.append(d[9])
b_chest.append(d[5])
b_chest.append(d[6])
b_chest.append(d[11])
b_chest.append(d[12])
Keypoints_left_arm_scores.append(np.asarray(b_left_arm))
Keypoints_chest_scores.append(np.asarray(b_chest))
Keypoints_right_arm_scores.append(np.asarray(b_right_arm))
for d in Keypoints_face_coords:
mean_points_face_coords.append(np.mean(d,axis=0))
for d in Keypoints_face_scores:
mean_points_face_scores.append(np.mean(d))
for d in Keypoints_chest_coords:
mean_points_chest_coords.append(np.mean(d,axis=0))
for d in Keypoints_chest_scores:
mean_points_chest_scores.append(np.mean(d))
for d in Keypoints_right_arm_coords:
mean_points_right_arm_coords.append(np.mean(d,axis=0))
for d in Keypoints_right_arm_scores:
mean_points_right_arm_scores.append(np.mean(d))
for d in Keypoints_left_arm_coords:
mean_points_left_arm_coords.append(np.mean(d,axis=0))
for d in Keypoints_left_arm_scores:
mean_points_left_arm_scores.append(np.mean(d))
for ii,d in enumerate(pose_scores):
means_prov_coords.append(mean_points_face_coords[ii])
means_prov_scores.append(mean_points_face_scores[ii])
#means_prov_scores.append(mean_points_chest_scores[ii])
#means_prov_coords.append(mean_points_chest_coords[ii])
#means_prov_scores.append(mean_points_right_arm_scores[ii])
#means_prov_coords.append(mean_points_right_arm_coords[ii])
#means_prov_scores.append(mean_points_left_arm_scores[ii])
#means_prov_coords.append(mean_points_left_arm_coords[ii])
total_means_coords.append(np.asarray(means_prov_coords))
total_means_scores.append(np.asarray(means_prov_scores))
Keypoints_face_scores = np.asarray(Keypoints_face_scores)
Keypoints_face_coords = np.asarray(Keypoints_face_coords)
Keypoints_right_arm_coords = np.asarray(Keypoints_right_arm_coords)
Keypoints_right_arm_scores = np.asarray(Keypoints_right_arm_scores)
Keypoints_left_arm_coords = np.asarray(Keypoints_left_arm_coords)
Keypoints_left_arm_scores = np.asarray(Keypoints_left_arm_scores)
total_means_coords = np.asarray(total_means_coords)
total_means_scores = np.asarray(total_means_scores)
Keypoints_att_show_coords = np.asarray(Keypoints_att_show_coords)
Keypoints_att_show_scores = np.asarray(Keypoints_att_show_scores)
overlay_image = posenet.draw_skel_and_kp( display_image, pose_scores, keypoint_scores, keypoint_coords, min_pose_score=0.15, min_part_score=0.1)
overlay_image_keypoints = posenet.draw_keypoints(display_image, pose_scores, keypoint_scores, keypoint_coords, min_pose_score=0.15, min_part_score=0.15)
overlay_image_face = posenet.draw_face(display_image, pose_scores, Keypoints_face_scores, Keypoints_face_coords, min_pose_score=0.15, min_part_score=0.15)
overlay_image_right_arm = posenet.draw_arm_right(display_image, pose_scores, Keypoints_right_arm_scores, Keypoints_right_arm_coords,min_pose_score=0.15, min_part_score=0.15)
overlay_image_left_arm = posenet.draw_arm_left(display_image, pose_scores, Keypoints_left_arm_scores, Keypoints_left_arm_coords,min_pose_score=0.15, min_part_score=0.15)
overlay_image_chest = posenet.draw_chest(display_image, pose_scores, Keypoints_chest_scores, Keypoints_chest_coords, min_pose_score=0.15, min_part_score=0.15)
overlay_image_means = posenet.draw_means(display_image, pose_scores, total_means_scores,total_means_coords,min_pose_score=0.15, min_part_score=0.15)
overlay_image_att = posenet.draw_att(display_image, pose_scores, Keypoints_att_show_scores,Keypoints_att_show_coords,min_pose_score=0.15, min_part_score=0.15)
overlay_image_skeleton = posenet.draw_skeleton(display_image, pose_scores, keypoint_scores, keypoint_coords,min_pose_score=0.15, min_part_score=0.15)
for idx,esq in enumerate(keypoint_coords):
x_coords = []
y_coords = []
for ii in range(0,min(args.detection_zone,len(esq)-1)):
if esq[ii][0]!=0 or esq[ii][1]!=0:
x_coords.append(esq[ii][0])
y_coords.append(esq[ii][1])
if len(x_coords)!=0:
x_min = np.min(x_coords)-20
y_min = np.min(y_coords)-20
x_max = np.max(x_coords)+20
y_max = np.max(y_coords)+20
x= x_min
y= y_min
w= x_max - x_min
h= y_max - y_min
x_bar= x + 0.5 * w
y_bar= y + 0.5 * h
matchedFid = None
for fid in faceTrackers.keys():
tracked_position = faceTrackers[fid].get_position()
t_x= int(tracked_position.left())
t_y= int(tracked_position.top())
t_w= int(tracked_position.width())
t_h= int(tracked_position.height())
t_x_bar= t_x + 0.5 * t_w
t_y_bar= t_y + 0.5 * t_h
if ((t_y<=x_bar<=(t_y+t_h))and(t_x<=y_bar<=(t_x+t_w))and(x<=t_y_bar<=(x+w))and(y<=t_x_bar<=(y+h))):
matchedFid = fid
centers=listofcenters[fid]
centers=[(int(y_bar),int(x_bar))]+centers
listofcenters[fid]=centers
max_distance=max_displacement[fid]
for (x,y) in centers:
distance=abs((pow(x_bar,2)+pow(y_bar,2))-(pow(t_x_bar,2)+pow(t_y_bar,2)))
max_displacement[fid]+=distance
if distance > max_movement_single[fid]:
max_movement_single[fid]=distance
if ((matchedFid is None)):
#print("Creating new tracker " + str(currentFaceID))
tracker = dlib.correlation_tracker()
tracker.start_track(overlay_image,dlib.rectangle(int(y_min), int(x_min), int(y_max) , int(x_max)))
faceTrackers[ currentFaceID ] = tracker
currentFaceID += 1
centers=[]
centers=[(int(y_bar),int(x_bar))]+centers
listofcenters.append(centers)
max_displacement.append(0)
max_movement_single.append(0)
for fid in faceTrackers.keys():
tracked_position = faceTrackers[fid].get_position()
t_x = int(tracked_position.left())
t_y = int(tracked_position.top())
t_w = int(tracked_position.width())
t_h = int(tracked_position.height())
t_x_bar = int(t_x + 0.5 * t_w)
t_y_bar = int(t_y + 0.5 * t_h)
centers=listofcenters[fid]
centers=[(t_x_bar,t_y_bar)]+centers
listofcenters[fid]=centers
max_distance=max_displacement[fid]
for (x,y) in centers:
distance=abs((pow(x_bar,2)+pow(y_bar,2))-(pow(t_x_bar,2)+pow(t_y_bar,2)))
max_displacement[fid]+=distance
if distance > max_movement_single[fid]:
max_movement_single[fid]=distance
cv2.rectangle(overlay_image, (int(t_x), int(t_y)),(int(t_x + t_w) ,int(t_y +t_h)), (0,255,0) ,2)
cv2.circle(overlay_image, (int(t_x_bar), int(t_y_bar)),5, (0,255,0) ,2)
# List to hold x values.
for fid in faceTrackers.keys():
values=listofcenters[fid]
stop=0
while stop < (len(values)-2):
x_1=int(values[stop][0])
y_1=int(values[stop][1])
x_2=int(values[stop+1][0])
y_2=int(values[stop+1][1])
# Plot the number in the list and set the line thickness.
if stop<10:
cv2.line(overlay_image,(x_1,y_1),(x_2,y_2),(255,0,0),4)
elif stop < 40:
cv2.line(overlay_image,(x_1,y_1),(x_2,y_2),(0,255,0),2)
elif stop < 100:
cv2.line(overlay_image,(x_1,y_1),(x_2,y_2),(0,0,255),1)
stop+=1
cv2.imshow('posenet', overlay_image)
#cv2.imshow('posenet_face', overlay_image_face)
#cv2.imshow('posenet_keypoints', overlay_image_keypoints)
#cv2.imshow('posenet_skeleton', overlay_image_skeleton)
#cv2.imshow('posenet_right_arm', overlay_image_right_arm)
#cv2.imshow('posenet_left_arm', overlay_image_left_arm)
#cv2.imshow('posenet_means', overlay_image_means)
#cv2.imshow('posenet_chest', overlay_image_chest)
#cv2.imshow('posenet_att', overlay_image_att)
#out.write(overlay_image)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
print('Average FPS: ', frame_count / (time.time() - start))
print('time: ', (time.time() - start))
cont=print_move(cont,max_displacement,max_movement_single)
def main():
with tf.Session() as sess:
model_cfg, model_outputs = posenet.load_model(args.model, sess)
output_stride = model_cfg['output_stride']
if args.file is not None:
cap = cv2.VideoCapture(args.file)
else:
cap = cv2.VideoCapture(args.cam_id)
cap.set(3, args.cam_width)
cap.set(4, args.cam_height)
i = 0
start = time.time()
frame_count = 0
left_center_cor = (0, 0)
right_center_cor = (0, 0)
while True:
input_image, display_image, output_scale = posenet.read_cap(
cap,
scale_factor=args.scale_factor,
output_stride=output_stride)
heatmaps_result, offsets_result, displacement_fwd_result, displacement_bwd_result = sess.run(
model_outputs, feed_dict={'image:0': input_image})
pose_scores, keypoint_scores, keypoint_coords = posenet.decode_multi.decode_multiple_poses(
heatmaps_result.squeeze(axis=0),
offsets_result.squeeze(axis=0),
displacement_fwd_result.squeeze(axis=0),
displacement_bwd_result.squeeze(axis=0),
output_stride=output_stride,
max_pose_detections=10,
min_pose_score=0.15)
keypoint_coords *= output_scale
overlay_image = posenet.draw_keypoints(display_image, pose_scores,
keypoint_scores[:, 9:11],
keypoint_coords[:, 9:11, :],
0.15, 0.1)
if i == 100:
i = 0
offset = random.randint(-50, 50)
left_center_cor = (int(keypoint_coords[:, 0, :][0][1]) + 100,
int(keypoint_coords[:, 0, :][0][0]) +
offset)
right_center_cor = (int(keypoint_coords[:, 1, :][0][1] - 100),
int(keypoint_coords[:, 1, :][0][0]) -
offset)
out_img = cv2.circle(overlay_image, left_center_cor, 10,
(0, 0, 255), -1)
out_img = cv2.circle(out_img, right_center_cor, 10, (0, 255, 255),
-1)
i += 1
cv2.imshow('Basket ball workout app', out_img)
frame_count += 1
if cv2.waitKey(1) & 0xFF == ord('q'):
break
print('Average FPS: ', frame_count / (time.time() - start))
CENTER_Y = CAMERA_RESOLUTION_HEIGHT//2
# MAIN LOOP
# ---------
while True:
success, img = cap.read() # read webcam capture
# get keypoints for single pose estimation. it is a list of 17 keypoints
keypoints = posenet.predict_singlepose(img)
# track nose
nose_pos = keypoints[0]['position']
nose_x = nose_pos[0] - CENTER_X
nose_y = nose_pos[1] - CENTER_Y
cv2.putText(img,f'x_distance:{nose_x} y_distance:{nose_y}', (0,CENTER_Y), cv2.FONT_HERSHEY_SIMPLEX, 1, (255,255,255), 2)
# draw keypoints to the original image
threshold = 0.0
draw_pose(img, keypoints, threshold=threshold)
draw_keypoints(img, keypoints, threshold=threshold)
poses = detect_pose(keypoints)
detected_poses = [pose for pose, detected in poses.items() if detected]
detected_poses = ' '.join(detected_poses) if detected_poses else 'None'
cv2.putText(img, f'{detected_poses} detected', (0,300), cv2.FONT_HERSHEY_SIMPLEX, 1, (255,255,255), 2)
cv2.imshow("pose", img) # show the image with keypoints
if cv2.waitKey(1) & 0xFF == ord('q'): # terminate window when press q
break
cv2.destroyAllWindows()
def main():
with tf.Session() as sess:
model_cfg, model_outputs = posenet.load_model(args.model, sess)
output_stride = model_cfg['output_stride']
cap = cv2.VideoCapture("data/Dabadaba/Cam_1/1_1.mp4")
cap.set(3, args.cam_width)
cap.set(4, args.cam_height)
len_video = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
fourcc = cv2.VideoWriter_fourcc(*'XVID')
output_movie = cv2.VideoWriter('data/track_video/pose_detection.avi',
fourcc, 29.97,
(args.cam_width, args.cam_height))
listofcenters = []
centers = []
max_displacement = []
max_movement_single = []
frame_number = 0
currentFaceID = 0
cont = 0
rectangleColor = (0, 0, 255)
faceTrackers = {}
start = time.time()
frame_count = 0
while True:
if int(time.time() - start) > ((cont + 1) * 5):
cont += 1
print("Total movement for the past {} seconds:".format(cont *
5))
for item in max_displacement:
print(item)
print("Maximun movement for the past {} seconds:".format(cont *
5))
for item_2 in max_movement_single:
print(item_2)
if frame_count == len_video:
break
res, img = cap.read()
if not res:
break
input_image, display_image, output_scale = posenet.process_input(
img,
scale_factor=args.scale_factor,
output_stride=output_stride)
#input_image, display_image, output_scale = posenet.read_cap( cap, scale_factor=args.scale_factor, output_stride=output_stride)
frame_count += 1
fidsToDelete = []
for fid in faceTrackers.keys():
trackingQuality = faceTrackers[fid].update(overlay_image)
if trackingQuality < trackingQuality_threshold:
fidsToDelete.append(fid)
for fid in fidsToDelete:
print("Removing fid " + str(fid) + " from list of trackers")
faceTrackers.pop(fid, None)
if (frame_number % n_frames_to_detect) == 0:
heatmaps_result, offsets_result, displacement_fwd_result, displacement_bwd_result = sess.run(
model_outputs, feed_dict={'image:0': input_image})
pose_scores, keypoint_scores, keypoint_coords = posenet.decode_multi.decode_multiple_poses(
heatmaps_result.squeeze(axis=0),
offsets_result.squeeze(axis=0),
displacement_fwd_result.squeeze(axis=0),
displacement_bwd_result.squeeze(axis=0),
output_stride=output_stride,
max_pose_detections=10,
min_pose_score=0.15)
keypoint_coords *= output_scale
Keypoints_face_coords = []
Keypoints_face_scores = []
for d in keypoint_coords:
Keypoints_face_coords.append(d[0:5])
for d in keypoint_scores:
Keypoints_face_scores.append(d[0:5])
Keypoints_eyes_coords = []
Keypoints_eyes_scores = []
for d in keypoint_coords:
Keypoints_eyes_coords.append(d[1:3])
for d in keypoint_scores:
Keypoints_eyes_scores.append(d[1:3])
Keypoints_half_eyes_coords = []
Keypoints_half_eyes_scores = []
for d in Keypoints_eyes_coords:
Keypoints_half_eyes_coords.append(np.mean(d, axis=0))
for d in Keypoints_eyes_scores:
Keypoints_half_eyes_scores.append(np.mean(d, axis=0))
Keypoints_angle_coords = []
Keypoints_angle_scores = []
angle = 0
for d in Keypoints_eyes_coords:
angle = math.atan2(d[1][1] - d[0][1], d[1][0] - d[0][0])
Keypoints_angle_coords.append(angle)
for d in Keypoints_eyes_scores:
Keypoints_angle_scores.append(np.mean(d, axis=0))
Keypoints_direction = []
angle_2 = 0
for ii in range(0, len(keypoint_coords) - 1):
if Keypoints_eyes_coords[ii][0][
0] == 0 and Keypoints_eyes_coords[ii][0][
1] == 0 and Keypoints_eyes_coords[ii][1][
0] == 0 and Keypoints_eyes_coords[ii][1][
1] == 0:
Keypoints_direction.append("no hay ojos") #rectoº
continue
elif Keypoints_eyes_coords[ii][0][
0] == 0 and Keypoints_eyes_coords[ii][0][1] == 0:
Keypoints_direction.append(
"ojo izquierdo (Derecha)") #rectoº
continue
elif Keypoints_eyes_coords[ii][1][
0] == 0 and Keypoints_eyes_coords[ii][1][1] == 0:
Keypoints_direction.append(
"ojo derecho (Izquierda)") #rectoº
continue
if Keypoints_face_coords[ii][0][
1] == 0 and Keypoints_face_coords[ii][0][
0] == 0 and Keypoints_half_eyes_coords[ii][
0] == 0 and Keypoints_half_eyes_coords[ii][
1] == 0:
Keypoints_direction.append("fallo") #rectoº
continue
angle_2 = math.atan2(
Keypoints_face_coords[ii][0][1] -
Keypoints_half_eyes_coords[ii][1],
Keypoints_face_coords[ii][0][0] -
Keypoints_half_eyes_coords[ii][0])
if abs(angle_2) < 0.261799:
Keypoints_direction.append("recto") #rectoº
elif angle_2 > 0.261799:
Keypoints_direction.append("derecha") #derecha
elif angle_2 < -0.261799:
Keypoints_direction.append("izquierda") #izq
Keypoints_att_show_coords = []
Keypoints_att_show_scores = []
a = []
b = []
for ii in range(0, len(keypoint_coords)):
print(ii)
a.append(Keypoints_face_coords[ii][1])
a.append(Keypoints_face_coords[ii][2])
a.append(Keypoints_face_coords[ii][0])
a.append(Keypoints_half_eyes_coords[ii])
Keypoints_att_show_coords.append(np.asarray(a))
for ii in range(0, len(keypoint_coords)):
b.append(Keypoints_face_scores[ii][1])
b.append(Keypoints_face_scores[ii][2])
b.append(Keypoints_face_scores[ii][0])
b.append(Keypoints_half_eyes_scores[ii])
Keypoints_att_show_scores.append(np.asarray(b))
Keypoints_right_arm_coords = []
Keypoints_right_arm_scores = []
a = []
b = []
for d in keypoint_coords:
a.append(d[6])
a.append(d[8])
a.append(d[10])
Keypoints_right_arm_coords.append(np.asarray(a))
for d in keypoint_scores:
b.append(d[6])
b.append(d[8])
b.append(d[10])
Keypoints_right_arm_scores.append(np.asarray(b))
Keypoints_left_arm_coords = []
Keypoints_left_arm_scores = []
a = []
b = []
for d in keypoint_coords:
a.append(d[5])
a.append(d[7])
a.append(d[9])
Keypoints_left_arm_coords.append(np.asarray(a))
for d in keypoint_scores:
b.append(d[5])
b.append(d[7])
b.append(d[9])
Keypoints_left_arm_scores.append(np.asarray(b))
Keypoints_chest_coords = []
Keypoints_chest_scores = []
a = []
b = []
for d in keypoint_coords:
a.append(d[5])
a.append(d[6])
a.append(d[11])
a.append(d[12])
Keypoints_chest_coords.append(np.asarray(a))
for d in keypoint_scores:
b.append(d[5])
b.append(d[6])
b.append(d[11])
b.append(d[12])
Keypoints_chest_scores.append(np.asarray(b))
mean_points_face_coords = []
mean_points_face_scores = []
mean_points_chest_coords = []
mean_points_chest_scores = []
mean_points_right_arm_coords = []
mean_points_right_arm_scores = []
mean_points_left_arm_coords = []
mean_points_left_arm_scores = []
for d in Keypoints_face_coords:
mean_points_face_coords.append(np.mean(d, axis=0))
for d in Keypoints_face_scores:
mean_points_face_scores.append(np.mean(d))
for d in Keypoints_chest_coords:
mean_points_chest_coords.append(np.mean(d, axis=0))
for d in Keypoints_chest_scores:
mean_points_chest_scores.append(np.mean(d))
for d in Keypoints_right_arm_coords:
mean_points_right_arm_coords.append(np.mean(d, axis=0))
for d in Keypoints_right_arm_scores:
mean_points_right_arm_scores.append(np.mean(d))
for d in Keypoints_left_arm_coords:
mean_points_left_arm_coords.append(np.mean(d, axis=0))
for d in Keypoints_left_arm_scores:
mean_points_left_arm_scores.append(np.mean(d))
total_means_coords = []
means_prov_coords = []
for ii, d in enumerate(pose_scores):
means_prov_coords.append(mean_points_face_coords[ii])
#means_prov_coords.append(mean_points_chest_coords[ii])
#means_prov_coords.append(mean_points_right_arm_coords[ii])
#means_prov_coords.append(mean_points_left_arm_coords[ii])
total_means_coords.append(np.asarray(means_prov_coords))
total_means_scores = []
means_prov_scores = []
for ii, d in enumerate(pose_scores):
means_prov_scores.append(mean_points_face_scores[ii])
#means_prov_scores.append(mean_points_chest_scores[ii])
#means_prov_scores.append(mean_points_right_arm_scores[ii])
#means_prov_scores.append(mean_points_left_arm_scores[ii])
total_means_scores.append(np.asarray(means_prov_scores))
Keypoints_face_scores = np.asarray(Keypoints_face_scores)
Keypoints_face_coords = np.asarray(Keypoints_face_coords)
Keypoints_right_arm_coords = np.asarray(
Keypoints_right_arm_coords)
Keypoints_right_arm_scores = np.asarray(
Keypoints_right_arm_scores)
Keypoints_left_arm_coords = np.asarray(
Keypoints_left_arm_coords)
Keypoints_left_arm_scores = np.asarray(
Keypoints_left_arm_scores)
total_means_coords = np.asarray(total_means_coords)
total_means_scores = np.asarray(total_means_scores)
Keypoints_att_show_coords = np.asarray(
Keypoints_att_show_coords)
Keypoints_att_show_scores = np.asarray(
Keypoints_att_show_scores)
overlay_image = posenet.draw_skel_and_kp(display_image,
pose_scores,
keypoint_scores,
keypoint_coords,
min_pose_score=0.15,
min_part_score=0.1)
overlay_image_keypoints = posenet.draw_keypoints(
display_image,
pose_scores,
keypoint_scores,
keypoint_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_face = posenet.draw_face(display_image,
pose_scores,
Keypoints_face_scores,
Keypoints_face_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_right_arm = posenet.draw_arm_right(
display_image,
pose_scores,
Keypoints_right_arm_scores,
Keypoints_right_arm_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_left_arm = posenet.draw_arm_left(
display_image,
pose_scores,
Keypoints_left_arm_scores,
Keypoints_left_arm_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_chest = posenet.draw_chest(
display_image,
pose_scores,
Keypoints_chest_scores,
Keypoints_chest_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_means = posenet.draw_means(display_image,
pose_scores,
total_means_scores,
total_means_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_att = posenet.draw_att(display_image,
pose_scores,
Keypoints_att_show_scores,
Keypoints_att_show_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_skeleton = posenet.draw_skeleton(
display_image,
pose_scores,
keypoint_scores,
keypoint_coords,
min_pose_score=0.15,
min_part_score=0.15)
for idx, esq in enumerate(keypoint_coords):
x_coords = []
y_coords = []
for ii in range(0, min(min_points, len(esq) - 1)):
if esq[ii][0] != 0 or esq[ii][1] != 0:
x_coords.append(esq[ii][0])
y_coords.append(esq[ii][1])
if len(x_coords) != 0:
x_min = np.min(x_coords) - 20
y_min = np.min(y_coords) - 20
x_max = np.max(x_coords) + 20
y_max = np.max(y_coords) + 20
x = x_min
y = y_min
w = x_max - x_min
h = y_max - y_min
x_bar = x + 0.5 * w
y_bar = y + 0.5 * h
#print(esq,x_min,x_max,y_max,y_min,"\n\n\n")
#cv2.rectangle(overlay_image, (int(y_min), int(x_min)),(int(y_max) ,int(x_max)), (0,255,0) ,2)
#cv2.circle(overlay_image, (int(y_bar), int(x_bar)),5, (0,255,0) ,2)
#cv2.circle(overlay_image, (int(y_max), int(x_max)),5, (0,255,0) ,2)
matchedFid = None
#
for fid in faceTrackers.keys():
tracked_position = faceTrackers[fid].get_position()
t_x = int(tracked_position.left())
t_y = int(tracked_position.top())
t_w = int(tracked_position.width())
t_h = int(tracked_position.height())
t_x_bar = t_x + 0.5 * t_w
t_y_bar = t_y + 0.5 * t_h
#cv2.circle(overlay_image, (int(t_x_bar), int(t_y_bar)),5, (255,0,0) ,2)
#cv2.circle(overlay_image, (int(t_x_bar), int(t_x_bar)),5, (255,0,0) ,2)
if ((t_y <= x_bar <=
(t_y + t_h)) and (t_x <= y_bar <= (t_x + t_w))
and (x <= t_y_bar <= (x + w))
and (y <= t_x_bar <= (y + h))):
matchedFid = fid
centers = listofcenters[fid]
centers = [(int(y_bar), int(x_bar))] + centers
listofcenters[fid] = centers
max_distance = max_displacement[fid]
for (x, y) in centers:
distance = abs(
(pow(x_bar, 2) + pow(y_bar, 2)) -
(pow(t_x_bar, 2) + pow(t_y_bar, 2)))
max_displacement[fid] += distance
if distance > max_movement_single[fid]:
max_movement_single[fid] = distance
if ((matchedFid is None)):
print("Creating new tracker " + str(currentFaceID))
tracker = dlib.correlation_tracker()
tracker.start_track(
overlay_image,
dlib.rectangle(int(y_min), int(x_min),
int(y_max), int(x_max)))
faceTrackers[currentFaceID] = tracker
currentFaceID += 1
centers = []
centers = [(int(y_bar), int(x_bar))] + centers
listofcenters.append(centers)
max_displacement.append(0)
max_movement_single.append(0)
for fid in faceTrackers.keys():
tracked_position = faceTrackers[fid].get_position()
t_x = int(tracked_position.left())
t_y = int(tracked_position.top())
t_w = int(tracked_position.width())
t_h = int(tracked_position.height())
t_x_bar = int(t_x + 0.5 * t_w)
t_y_bar = int(t_y + 0.5 * t_h)
centers = listofcenters[fid]
centers = [(t_x_bar, t_y_bar)] + centers
listofcenters[fid] = centers
max_distance = max_displacement[fid]
for (x, y) in centers:
distance = abs((pow(x_bar, 2) + pow(y_bar, 2)) -
(pow(t_x_bar, 2) + pow(t_y_bar, 2)))
max_displacement[fid] += distance
if distance > max_movement_single[fid]:
max_movement_single[fid] = distance
cv2.rectangle(overlay_image, (int(t_x), int(t_y)),
(int(t_x + t_w), int(t_y + t_h)), (0, 255, 0), 2)
cv2.circle(overlay_image, (int(t_x_bar), int(t_y_bar)), 5,
(0, 255, 0), 2)
# List to hold x values.
for fid in faceTrackers.keys():
values = listofcenters[fid]
stop = 0
while stop < (len(values) - 2):
x_1 = int(values[stop][0])
y_1 = int(values[stop][1])
x_2 = int(values[stop + 1][0])
y_2 = int(values[stop + 1][1])
# Plot the number in the list and set the line thickness.
if stop < 10:
cv2.line(overlay_image, (x_1, y_1), (x_2, y_2),
(255, 0, 0), 4)
elif stop < 40:
cv2.line(overlay_image, (x_1, y_1), (x_2, y_2),
(0, 255, 0), 2)
elif stop < 100:
cv2.line(overlay_image, (x_1, y_1), (x_2, y_2),
(0, 0, 255), 1)
stop += 1
#cv2.imshow('posenet', overlay_image)
#cv2.imshow('posenet_face', overlay_image_face)
#cv2.imshow('posenet_keypoints', overlay_image_keypoints)
#cv2.imshow('posenet_skeleton', overlay_image_skeleton)
#cv2.imshow('posenet_right_arm', overlay_image_right_arm)
#cv2.imshow('posenet_left_arm', overlay_image_left_arm)
#cv2.imshow('posenet_means', overlay_image_means)
#cv2.imshow('posenet_chest', overlay_image_chest)
#cv2.imshow('posenet_att', overlay_image_att)
#output_movie.write(overlay_image)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
print('Average FPS: ', frame_count / (time.time() - start))
print('time: ', (time.time() - start))
print("Total movement for the past {} seconds:".format(time.time()))
for item in max_displacement:
print(item)
print("Maximun movement for the past {} seconds:".format(time.time()))
for item_2 in max_movement_single:
print(item_2)
def main():
with tf.Session() as sess:
model_cfg, model_outputs = posenet.load_model(args.model, sess)
output_stride = model_cfg['output_stride']
cap = cv2.VideoCapture("data/Dabadaba/Cam_1/2_1.mp4")
frame_width = int(cap.get(3))
frame_height = int(cap.get(4))
cap.set(3, args.cam_width)
cap.set(4, args.cam_height)
len_video = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
fourcc = cv2.VideoWriter_fourcc(*'XVID')
output_movie = cv2.VideoWriter('data/daba_out/pose_detection_2_1.avi',
fourcc, 29, (frame_width, frame_height))
start = time.time()
frame_count = 0
while True:
if frame_count == len_video - 1:
break
res, img = cap.read()
if not res:
break
input_image, display_image, output_scale = posenet.process_input(
img,
scale_factor=args.scale_factor,
output_stride=output_stride)
#input_image, display_image, output_scale = posenet.read_cap( cap, scale_factor=args.scale_factor, output_stride=output_stride)
heatmaps_result, offsets_result, displacement_fwd_result, displacement_bwd_result = sess.run(
model_outputs, feed_dict={'image:0': input_image})
pose_scores, keypoint_scores, keypoint_coords = posenet.decode_multi.decode_multiple_poses(
heatmaps_result.squeeze(axis=0),
offsets_result.squeeze(axis=0),
displacement_fwd_result.squeeze(axis=0),
displacement_bwd_result.squeeze(axis=0),
output_stride=output_stride,
max_pose_detections=50,
min_pose_score=0.15)
keypoint_coords *= output_scale
Keypoints_face_coords = []
Keypoints_face_scores = []
for d in keypoint_coords:
Keypoints_face_coords.append(d[0:5])
for d in keypoint_scores:
Keypoints_face_scores.append(d[0:5])
Keypoints_eyes_coords = []
Keypoints_eyes_scores = []
for d in keypoint_coords:
Keypoints_eyes_coords.append(d[1:3])
for d in keypoint_scores:
Keypoints_eyes_scores.append(d[1:3])
Keypoints_half_eyes_coords = []
Keypoints_half_eyes_scores = []
for d in Keypoints_eyes_coords:
Keypoints_half_eyes_coords.append(np.mean(d, axis=0))
for d in Keypoints_eyes_scores:
Keypoints_half_eyes_scores.append(np.mean(d, axis=0))
Keypoints_angle_coords = []
Keypoints_angle_scores = []
angle = 0
for d in Keypoints_eyes_coords:
angle = math.atan2(d[1][1] - d[0][1], d[1][0] - d[0][0])
Keypoints_angle_coords.append(angle)
for d in Keypoints_eyes_scores:
Keypoints_angle_scores.append(np.mean(d, axis=0))
Keypoints_direction = []
angle_2 = 0
for ii in range(0, len(keypoint_coords) - 1):
if Keypoints_eyes_coords[ii][0][
0] == 0 and Keypoints_eyes_coords[ii][0][
1] == 0 and Keypoints_eyes_coords[ii][1][
0] == 0 and Keypoints_eyes_coords[ii][1][
1] == 0:
Keypoints_direction.append("no hay ojos") #rectoº
continue
elif Keypoints_eyes_coords[ii][0][
0] == 0 and Keypoints_eyes_coords[ii][0][1] == 0:
Keypoints_direction.append(
"ojo izquierdo (Derecha)") #rectoº
continue
elif Keypoints_eyes_coords[ii][1][
0] == 0 and Keypoints_eyes_coords[ii][1][1] == 0:
Keypoints_direction.append(
"ojo derecho (Izquierda)") #rectoº
continue
if Keypoints_face_coords[ii][0][
1] == 0 and Keypoints_face_coords[ii][0][
0] == 0 and Keypoints_half_eyes_coords[ii][
0] == 0 and Keypoints_half_eyes_coords[ii][
1] == 0:
Keypoints_direction.append("fallo") #rectoº
continue
angle_2 = math.atan2(
Keypoints_face_coords[ii][0][1] -
Keypoints_half_eyes_coords[ii][1],
Keypoints_face_coords[ii][0][0] -
Keypoints_half_eyes_coords[ii][0])
if abs(angle_2) < 0.261799:
Keypoints_direction.append("recto") #rectoº
elif angle_2 > 0.261799:
Keypoints_direction.append("derecha") #derecha
elif angle_2 < -0.261799:
Keypoints_direction.append("izquierda") #izq
Keypoints_att_show_coords = []
Keypoints_att_show_scores = []
a = []
b = []
for ii in range(0, len(keypoint_coords) - 1):
a.append(Keypoints_face_coords[ii][1])
a.append(Keypoints_face_coords[ii][2])
a.append(Keypoints_face_coords[ii][0])
a.append(Keypoints_half_eyes_coords[ii])
Keypoints_att_show_coords.append(np.asarray(a))
for ii in range(0, len(keypoint_coords) - 1):
b.append(Keypoints_face_scores[ii][1])
b.append(Keypoints_face_scores[ii][2])
b.append(Keypoints_face_scores[ii][0])
b.append(Keypoints_half_eyes_scores[ii])
Keypoints_att_show_scores.append(np.asarray(b))
Keypoints_right_arm_coords = []
Keypoints_right_arm_scores = []
a = []
b = []
for d in keypoint_coords:
a.append(d[6])
a.append(d[8])
a.append(d[10])
Keypoints_right_arm_coords.append(np.asarray(a))
for d in keypoint_scores:
b.append(d[6])
b.append(d[8])
b.append(d[10])
Keypoints_right_arm_scores.append(np.asarray(b))
Keypoints_left_arm_coords = []
Keypoints_left_arm_scores = []
a = []
b = []
for d in keypoint_coords:
a.append(d[5])
a.append(d[7])
a.append(d[9])
Keypoints_left_arm_coords.append(np.asarray(a))
for d in keypoint_scores:
b.append(d[5])
b.append(d[7])
b.append(d[9])
Keypoints_left_arm_scores.append(np.asarray(b))
Keypoints_chest_coords = []
Keypoints_chest_scores = []
a = []
b = []
for d in keypoint_coords:
a.append(d[5])
a.append(d[6])
a.append(d[11])
a.append(d[12])
Keypoints_chest_coords.append(np.asarray(a))
for d in keypoint_scores:
b.append(d[5])
b.append(d[6])
b.append(d[11])
b.append(d[12])
Keypoints_chest_scores.append(np.asarray(b))
mean_points_face_coords = []
mean_points_face_scores = []
mean_points_chest_coords = []
mean_points_chest_scores = []
mean_points_right_arm_coords = []
mean_points_right_arm_scores = []
mean_points_left_arm_coords = []
mean_points_left_arm_scores = []
for d in Keypoints_face_coords:
mean_points_face_coords.append(np.mean(d, axis=0))
for d in Keypoints_face_scores:
mean_points_face_scores.append(np.mean(d))
for d in Keypoints_chest_coords:
mean_points_chest_coords.append(np.mean(d, axis=0))
for d in Keypoints_chest_scores:
mean_points_chest_scores.append(np.mean(d))
for d in Keypoints_right_arm_coords:
mean_points_right_arm_coords.append(np.mean(d, axis=0))
for d in Keypoints_right_arm_scores:
mean_points_right_arm_scores.append(np.mean(d))
for d in Keypoints_left_arm_coords:
mean_points_left_arm_coords.append(np.mean(d, axis=0))
for d in Keypoints_left_arm_scores:
mean_points_left_arm_scores.append(np.mean(d))
total_means_coords = []
means_prov_coords = []
for ii, d in enumerate(pose_scores):
means_prov_coords.append(mean_points_face_coords[ii])
#means_prov_coords.append(mean_points_chest_coords[ii])
#means_prov_coords.append(mean_points_right_arm_coords[ii])
#means_prov_coords.append(mean_points_left_arm_coords[ii])
total_means_coords.append(np.asarray(means_prov_coords))
total_means_scores = []
means_prov_scores = []
for ii, d in enumerate(pose_scores):
means_prov_scores.append(mean_points_face_scores[ii])
#means_prov_scores.append(mean_points_chest_scores[ii])
#means_prov_scores.append(mean_points_right_arm_scores[ii])
#means_prov_scores.append(mean_points_left_arm_scores[ii])
total_means_scores.append(np.asarray(means_prov_scores))
Keypoints_face_scores = np.asarray(Keypoints_face_scores)
Keypoints_face_coords = np.asarray(Keypoints_face_coords)
Keypoints_right_arm_coords = np.asarray(Keypoints_right_arm_coords)
Keypoints_right_arm_scores = np.asarray(Keypoints_right_arm_scores)
Keypoints_left_arm_coords = np.asarray(Keypoints_left_arm_coords)
Keypoints_left_arm_scores = np.asarray(Keypoints_left_arm_scores)
total_means_coords = np.asarray(total_means_coords)
total_means_scores = np.asarray(total_means_scores)
Keypoints_att_show_coords = np.asarray(Keypoints_att_show_coords)
Keypoints_att_show_scores = np.asarray(Keypoints_att_show_scores)
overlay_image_keypoints = posenet.draw_keypoints(
display_image,
pose_scores,
keypoint_scores,
keypoint_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image = posenet.draw_skel_and_kp(display_image,
pose_scores,
keypoint_scores,
keypoint_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_face = posenet.draw_face(display_image,
pose_scores,
Keypoints_face_scores,
Keypoints_face_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_right_arm = posenet.draw_arm_right(
display_image,
pose_scores,
Keypoints_right_arm_scores,
Keypoints_right_arm_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_left_arm = posenet.draw_arm_left(
display_image,
pose_scores,
Keypoints_left_arm_scores,
Keypoints_left_arm_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_chest = posenet.draw_chest(display_image,
pose_scores,
Keypoints_chest_scores,
Keypoints_chest_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_means = posenet.draw_means(display_image,
pose_scores,
total_means_scores,
total_means_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_att = posenet.draw_att(display_image,
pose_scores,
Keypoints_att_show_scores,
Keypoints_att_show_coords,
min_pose_score=0.15,
min_part_score=0.15)
overlay_image_skeleton = posenet.draw_skeleton(display_image,
pose_scores,
keypoint_scores,
keypoint_coords,
min_pose_score=0.15,
min_part_score=0.15)
#cv2.imshow('posenet_body', overlay_image)
#cv2.imshow('posenet_face', overlay_image_face)
#cv2.imshow('posenet_keypoints', overlay_image_keypoints)
#cv2.imshow('posenet_skeleton', overlay_image_skeleton)
#cv2.imshow('posenet_right_arm', overlay_image_right_arm)
#cv2.imshow('posenet_left_arm', overlay_image_left_arm)
#cv2.imshow('posenet_means', overlay_image_means)
#cv2.imshow('posenet_chest', overlay_image_chest)
cv2.imshow('posenet_att', overlay_image_att)
#output_movie.write(overlay_image)
frame_count += 1
if cv2.waitKey(1) & 0xFF == ord('q'):
break
print('Average FPS: ', frame_count / (time.time() - start))
print('time: ', (time.time() - start))