def gen_cropped_frames(video_path, coords_path, save_path):
cap = cv2.VideoCapture(video_path)
fname = video_path.split('/')[-1].split('.')[0]
fnum = fname.split("_")[-1]
frame_count = 0
coords = open(coords_path).readlines()
total_frames = int(cap.get(cv2.cv.CV_CAP_PROP_FRAME_COUNT))
if (total_frames-len(coords)) > 0:
append_frame_skip = int(total_frames/(total_frames-len(coords)))
for f in range(0,total_frames, append_frame_skip):
coords.insert(f, coords[f])
vid = my_video_capture(video_path, frame_rate=30)
crop_coords_used = open("%s/%s/cropped/crop_coords.txt" % (save_path, fname), "wb")
output_vid = my_video_capture(save_path, frame_rate=30, mode="write")
prev_fname = fname.split("_")[:-1]
prev_fname.append(str(int(fnum) - 1).zfill(4))
prev_fname = "_".join(prev_fname)
if os.path.exists("%s/%s/cropped/crop_coords.txt" % (save_path, prev_fname)):
use_coord = np.array([int(n) for n in open("%s/%s/cropped/crop_coords.txt" % (save_path, prev_fname)).readlines()[-1].split(",")])
else:
use_coord = np.array([0,0,0,0])
while vid.has_next():
frame = vid.read()
try:
cur_coord = np.array([int(n) for n in coords[frame_count].split(",")])
future_coords = [np.array([int(n) for n in coords[frame_count+f_offset].split(",")]) for f_offset in xrange(5)]
except IndexError:
pdb.set_trace()
diff = np.mean(np.sum(np.abs(future_coord-use_coord)) for future_coord in future_coords)
if diff > 100:
use_coord = cur_coord
crop_coords_used.write(",".join([str(i) for i in use_coord]) + "\n")
frame_count += 1
if sum(use_coord) > 0:
output_vid.write(frame[use_coord[2]:use_coord[3], use_coord[0]:use_coord[1]])
else:
output_vid.write(frame)
output_vid.new_img_folder("%s/%s/cropped/" % (save_path, fname))
output_vid.close()
vid.close()
def main(sbj_id, day, result, video_loc, disconnect_file, save_fldr, comp):
img_fldr = "%s/images" % save_fldr
img_fldr_final = "%s/images_final" % save_fldr
if not os.path.exists(img_fldr):
os.makedirs(img_fldr)
if not os.path.exists(img_fldr_final):
os.makedirs(img_fldr_final)
start_time, end_time, start, end = get_disconnected_times(disconnect_file)
duration = 60*60*1
video_starts = pickle.load(open("C:\\Users\\wangnxr\\Documents\\rao_lab\\video_analysis\\vid_real_time\\%s_%i.p" % (sbj_id, day), "rb"))
#plot_2d_coords(result, (4,9), (25,55), img_fldr, day, start_time, duration, type="slow")
for i, video_start in enumerate(video_starts["start"]):
if (start_time+timedelta(seconds=150))<=video_start:
vid_counter=i-1
break
frame_counter=(start_time+timedelta(seconds=150)-video_starts["start"][vid_counter]).seconds+1
cur_vid = my_video_capture("%s/%s_%i_%04i.avi" %(video_loc, sbj_id, day, vid_counter), 1)
movie = my_video_capture(img_fldr_final, 30, mode="write")
for s in range(0,duration-150,1):
plot = cv2.resize(cv2.imread("%s/%i_%05d.png" % (img_fldr, day, s)), (480, 480))
cur_time = start_time + timedelta(seconds=(s+150))
if not disconnected(start, end, cur_time):
if vid_counter < len(video_starts["start"]) and cur_time >= video_starts["start"][vid_counter+1]:
vid_counter += 1
frame_counter = 1
cur_vid.close()
cur_vid = my_video_capture("%s/%s_%i_%04i.avi" %(video_loc, sbj_id, day, vid_counter), 1)
cur_vid.forward_to(frame_counter)
frame = cur_vid.read()
frame_counter += 1
else:
frame = np.zeros((480,640,3), np.uint8)
cv2.putText(frame, "Shown at 30X", (50,50), cv2.FONT_HERSHEY_SIMPLEX, 1, (255,255,255))
concat = concat_images(plot, frame)
movie.write(concat)
cur_vid.close()
movie.new_vid("%s/movie_comp%i_for_bing3.avi" % (save_fldr, comp))
movie.close()
def face_blur(vid_src):
#cascade_fn = "C:\\Python27\\lib\\site-packages\\pyESig\\vid\\haarcascades\\haarcascade_frontalface_alt.xml"
#cascade = cv2.CascadeClassifier(cascade_fn)
cam = my_video_capture(vid_src, 30)
frame_cnt = 1
while cam.has_next():
img = cam.read()
height, width, depth = img.shape
#gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
#gray = cv2.equalizeHist(gray)
min_neigh = 1
#t = clock()
# if frame_cnt%90==1:
# rects = detect(gray[50:-50,50:-50], cascade, min_neigh)
# rects = []
# # while len(rects) == 0 and cam.has_next():
# # img_tmp = cam.read()
# # height, width, depth = img_tmp.shape
# # gray = cv2.cvtColor(img_tmp, cv2.COLOR_BGR2GRAY)
# # gray = cv2.equalizeHist(gray)
# # rects = detect(gray[50:-50,50:-50], cascade, min_neigh)
# cam.rewind_to(frame_cnt)
# img = cam.read()
# c_x = width/2
# c_y = height/2
# (x1,y1,x2,y2) = find_center_rect(rects, (c_x,c_y))
x1 = 0
x2 = 50
y1 = 60
y2 = 160
sub_face = img[y1:y2, x1:x2]
sub_face=cv2.GaussianBlur(sub_face, (23,23),15)
img[y1:y2, x1:x2] = sub_face
#cv2.imshow('facedetect', img)
#pdb.set_trace()
if 0xFF & cv2.waitKey(5) == 27:
break
cam.write(img, frame_cnt)
frame_cnt += 1
print frame_cnt
fileName, fileExt = os.path.splitext(vid_src)
cam.new_vid(fileName + "_2" + fileExt)
cam.close()
cv2.destroyAllWindows()
def main(joints_file, vid_file, save_dir):
poses = np.array(
[numerate_coords(row) for row in (open(joints_file)).readlines()])
downsample_fact = 6
difficult_frames = []
for p, pose in enumerate(poses):
if p % downsample_fact == 0 and (
len(difficult_frames) == 0 or
(p - difficult_frames[-1]) > 30) and np.any(pose[:, 2] < 0.1):
difficult_frames.append(p)
vid_name = os.path.split(vid_file)[-1].split('.')[0]
vid_file = my_video_capture(vid_file, 30 / downsample_fact)
for f in difficult_frames:
if f % downsample_fact == 0:
vid_file.forward_to(f / downsample_fact)
img = vid_file.read()
cv2.imwrite(os.path.join(save_dir, "%s_%i.jpg" % (vid_name, f)),
img)
def main(mv_file, vid_file, save_dir, offset):
mv_file = pickle.load(open(mv_file))
if np.sum(mv_file[np.where(mv_file > 0)]) < (0.5 * 3000):
#Very little going on in this file
print vid_file
return
vid_name = os.path.split(vid_file)[-1].split('.')[0]
vid_file = my_video_capture(vid_file, 30)
left_arm_mvmt = np.sum(mv_file[:, (2, 4, 6)], axis=1)
right_arm_mvmt = np.sum(mv_file[:, (1, 3, 5)], axis=1)
head_mvmt = mv_file[:, 0]
train_dir = os.path.join(save_dir, "train")
test_dir = os.path.join(save_dir, "test")
if not os.path.exists(train_dir):
os.makedirs(train_dir)
if not os.path.exists(test_dir):
os.makedirs(test_dir)
if not os.path.exists(os.path.join(train_dir, "l_arm_1")):
os.makedirs(os.path.join(train_dir, "head_0"))
os.makedirs(os.path.join(train_dir, "head_1"))
os.makedirs(os.path.join(train_dir, "r_arm_0"))
os.makedirs(os.path.join(train_dir, "r_arm_1"))
os.makedirs(os.path.join(train_dir, "l_arm_0"))
os.makedirs(os.path.join(train_dir, "l_arm_1"))
os.makedirs(os.path.join(train_dir, "mv_1"))
os.makedirs(os.path.join(train_dir, "mv_0"))
if not os.path.exists(os.path.join(test_dir, "l_arm_1")):
os.makedirs(os.path.join(test_dir, "head_0"))
os.makedirs(os.path.join(test_dir, "head_1"))
os.makedirs(os.path.join(test_dir, "r_arm_0"))
os.makedirs(os.path.join(test_dir, "r_arm_1"))
os.makedirs(os.path.join(test_dir, "l_arm_0"))
os.makedirs(os.path.join(test_dir, "l_arm_1"))
os.makedirs(os.path.join(test_dir, "mv_1"))
os.makedirs(os.path.join(test_dir, "mv_0"))
if int(day) == 9:
cur_dir = test_dir
else:
cur_dir = train_dir
for f in range(offset + 1 + 15, len(mv_file), 10):
vid_file.forward_to(f - offset)
img = vid_file.read()
if np.mean(left_arm_mvmt[f:f + 5]) > 2:
cv2.imwrite(
os.path.join(cur_dir, "l_arm_1",
"%s_%i.png" % (vid_name, f - offset)), img)
cv2.imwrite(
os.path.join(cur_dir, "mv_1",
"%s_%i.png" % (vid_name, f - offset)), img)
if np.mean(right_arm_mvmt[f:f + 5]) > 2:
cv2.imwrite(
os.path.join(cur_dir, "r_arm_1",
"%s_%i.png" % (vid_name, f - offset)), img)
cv2.imwrite(
os.path.join(cur_dir, "mv_1",
"%s_%i.png" % (vid_name, f - offset)), img)
if np.mean(head_mvmt[f:f + 5]) > 1:
cv2.imwrite(
os.path.join(cur_dir, "head_1",
"%s_%i.png" % (vid_name, f - offset)), img)
cv2.imwrite(
os.path.join(cur_dir, "mv_1",
"%s_%i.png" % (vid_name, f - offset)), img)
vid_file.rewind()
for f in range(offset + 1 + 15, len(mv_file), 60):
vid_file.forward_to(f - offset)
img = vid_file.read()
flag = 0
if np.all(left_arm_mvmt[f:f + 5] >= 0) and np.mean(
left_arm_mvmt[f:f + 5]) < 1:
cv2.imwrite(
os.path.join(cur_dir, "l_arm_0",
"%s_%i.png" % (vid_name, f - offset)), img)
flag += 1
if np.all(right_arm_mvmt[f:f + 5] >= 0) and np.mean(
right_arm_mvmt[f:f + 5]) < 1:
cv2.imwrite(
os.path.join(cur_dir, "r_arm_0",
"%s_%i.png" % (vid_name, f - offset)), img)
flag += 1
if np.all(
head_mvmt[f:f + 5] >= 0) and np.mean(head_mvmt[f:f + 5]) < 0.5:
cv2.imwrite(
os.path.join(cur_dir, "head_0",
"%s_%i.png" % (vid_name, f - offset)), img)
flag += 1
#print (np.mean(left_arm_mvmt[f:f + 5]), np.mean(right_arm_mvmt[f:f + 5]), np.mean(head_mvmt[f:f + 5]))
if flag == 3:
cv2.imwrite(
os.path.join(cur_dir, "mv_0",
"%s_%i.png" % (vid_name, f - offset)), img)
def main(mv_file, vid_file, save_dir, offset):
mv_file = pickle.load(open(mv_file))
if np.sum(mv_file[np.where(mv_file>0)])<0.5*3000:
#Very little going on in this file
print vid_file
return
vid_name = os.path.split(vid_file)[-1].split('.')[0]
vid_file = my_video_capture(vid_file, 30)
left_arm_mvmt = np.sum(mv_file[:,(2,4,6)], axis=1)
right_arm_mvmt = np.sum(mv_file[:,(1,3,5)], axis=1)
head_mvmt = mv_file[:,0]
train_dir = os.path.join(save_dir, "train")
test_dir = os.path.join(save_dir, "test")
if not os.path.exists(train_dir):
os.makedirs(train_dir)
if not os.path.exists(test_dir):
os.makedirs(test_dir)
if not os.path.exists(os.path.join(train_dir, "l_arm_1")):
os.makedirs(os.path.join(train_dir,"head_0"))
os.makedirs(os.path.join(train_dir,"head_1"))
os.makedirs(os.path.join(train_dir,"r_arm_0"))
os.makedirs(os.path.join(train_dir,"r_arm_1"))
os.makedirs(os.path.join(train_dir,"l_arm_0"))
os.makedirs(os.path.join(train_dir,"l_arm_1"))
os.makedirs(os.path.join(train_dir,"mv_1"))
os.makedirs(os.path.join(train_dir,"mv_0"))
if not os.path.exists(os.path.join(test_dir,"l_arm_1")):
os.makedirs(os.path.join(test_dir,"head_0"))
os.makedirs(os.path.join(test_dir,"head_1"))
os.makedirs(os.path.join(test_dir,"r_arm_0"))
os.makedirs(os.path.join(test_dir,"r_arm_1"))
os.makedirs(os.path.join(test_dir,"l_arm_0"))
os.makedirs(os.path.join(test_dir,"l_arm_1"))
os.makedirs(os.path.join(test_dir,"mv_1"))
os.makedirs(os.path.join(test_dir,"mv_0"))
if np.random.randint(100) < 75:
cur_dir = train_dir
else:
cur_dir = test_dir
for f in range(offset+1,len(mv_file), 10):
vid_file.forward_to(f - offset)
img = vid_file.read()
if np.mean(left_arm_mvmt[f:f+5]>5):
#cv2.imwrite(os.path.join(cur_dir, "l_arm_1", "%s_%i.png" %(vid_name,f - offset)), img)
cv2.imwrite(os.path.join(cur_dir, "mv_1", "%s_%i.png" % (vid_name, f - offset)), img)
elif np.mean(right_arm_mvmt[f:f+5]>5):
#cv2.imwrite(os.path.join(cur_dir, "r_arm_1", "%s_%i.png" % (vid_name, f - offset)), img)
cv2.imwrite(os.path.join(cur_dir, "mv_1", "%s_%i.png" % (vid_name, f - offset)), img)
elif np.mean(head_mvmt[f:f+5]>3):
#cv2.imwrite(os.path.join(cur_dir, "head_1", "%s_%i.png" % (vid_name, f - offset)), img)
cv2.imwrite(os.path.join(cur_dir, "mv_1", "%s_%i.png" % (vid_name, f - offset)), img)
vid_file.rewind()
for f in range(offset+1, len(mv_file), 60):
vid_file.forward_to(f-offset)
img = vid_file.read()
flag = 0
if np.all(left_arm_mvmt[f:f+5] >= 0) and np.mean(left_arm_mvmt[f:f + 5]) < 1:
#cv2.imwrite(os.path.join(cur_dir, "l_arm_0", "%s_%i.png" % (vid_name, f - offset)), img)
flag+=1
if np.all(right_arm_mvmt[f:f + 5] >= 0) and np.mean(right_arm_mvmt[f:f + 5]) < 1:
#cv2.imwrite(os.path.join(cur_dir, "r_arm_0", "%s_%i.png" % (vid_name, f - offset)), img)
flag+=1
if np.all(head_mvmt[f:f + 5] >= 0) and np.mean(head_mvmt[f:f + 5]) < 1:
#cv2.imwrite(os.path.join(cur_dir, "head_0", "%s_%i.png" % (vid_name, f - offset)), img)
flag+=1
if flag==3:
cv2.imwrite(os.path.join(cur_dir, "mv_0", "%s_%i.png" % (vid_name, f - offset)), img)
