GMC / sync_videos.py
linhmv's picture
Add dataset
c9e1b42
import glob
import os
import pandas as pd
import numpy as np
import shutil
import os.path as osp
import cv2
import subprocess
import re
from calibration_syncvideos import calibration_syncvideos
ALLOWED_EXTENSIONS = ['.jpg', '.jpeg', '.npy', '.png', '.pcd']
def visual_sync_output(output_dirs, devices_name):
cam_txts = ["CAM " + devicen + " {}" for devicen in devices_name]
size = (1920, 1080) # (1920, 1080)
video_size = (size[0] * 2, size[1] * 2) # (size[0] * 3 + 1080, size[1] * 2)
fourcc = cv2.VideoWriter_fourcc(*'MJPG')
all_imgs = []
def extract_number(filename):
base = os.path.basename(filename)
number = int(os.path.splitext(base)[0])
return number
for outputdir in output_dirs:
all_imgs.append(sorted(glob.glob(outputdir + "/*"), key=extract_number))
out = cv2.VideoWriter("visual.avi", fourcc, 28, video_size)
for frame in range(len(all_imgs[0])):
# if frame < 600:
# continue
# if frame > 900:
# break
frame0 = cv2.imread(all_imgs[0][frame])
frame1 = cv2.imread(all_imgs[1][frame])
frame2 = cv2.imread(all_imgs[2][frame])
frame3 = cv2.imread(all_imgs[3][frame])
cv2.putText(frame0, cam_txts[0].format(frame), org=(100, 100), fontFace=cv2.FONT_HERSHEY_SIMPLEX, fontScale=2,
color=(0, 0, 255), thickness=3)
cv2.putText(frame1, cam_txts[1].format(frame), org=(100, 100), fontFace=cv2.FONT_HERSHEY_SIMPLEX, fontScale=2,
color=(0, 0, 255), thickness=3)
cv2.putText(frame2, cam_txts[2].format(frame), org=(100, 100), fontFace=cv2.FONT_HERSHEY_SIMPLEX, fontScale=2,
color=(0, 0, 255), thickness=3)
cv2.putText(frame3, cam_txts[3].format(frame), org=(100, 100), fontFace=cv2.FONT_HERSHEY_SIMPLEX, fontScale=2,
color=(0, 0, 255), thickness=3)
row0 = np.column_stack((frame0, frame1))
row1 = np.column_stack((frame2, frame3))
imshow = np.row_stack((row0, row1))
scale_percent = 0.5
dim = (int(imshow.shape[1] * scale_percent), int(imshow.shape[0] * scale_percent))
resized = cv2.resize(imshow, dim, interpolation=cv2.INTER_AREA) # resize image
cv2.imshow('Image', resized)
cv2.waitKey(10)
out.write(imshow)
out.release()
def sync_all_frames(input_dirs, output_dirs):
# Create a dictionary to hold image timestamps for each directory
timestamps_dict = {}
# Define a function to check if the basename contains only digits
def is_numeric_filename(filename):
basename = os.path.splitext(os.path.basename(filename))[0]
return basename.isdigit()
for vid in input_dirs:
out_images = sorted(glob.glob(vid + "/*"))
# Filter the out_images to include only numeric filenames
numeric_images = filter(is_numeric_filename, out_images)
image_timestamps = list(map(lambda x: int(os.path.splitext(os.path.basename(x))[0]), numeric_images))
timestamps_dict[vid] = image_timestamps
# Threshold for matching timestamps
THRESHOLD_NS = 25
# Create a DataFrame for each directory's timestamps
data_frames = {}
for vid, timestamps in timestamps_dict.items():
if 'frame' in timestamps:
continue
df = pd.DataFrame({'t': timestamps, vid: list(map(str, timestamps))})
data_frames[vid] = df
# Start with the first DataFrame
merged_df = list(data_frames.values())[0]
# Merge each subsequent DataFrame
for vid, df in list(data_frames.items())[1:]:
merged_df = pd.merge_asof(merged_df, df, on='t', tolerance=THRESHOLD_NS, allow_exact_matches=True,
direction='nearest')
# Drop any rows with NaN values resulting from the merge
merged_df = merged_df.dropna()
# Drop the 't' column if it's no longer needed
merged_df = merged_df.drop('t', axis='columns')
# Reset the index for a clean DataFrame
merged_df = merged_df.reset_index(drop=True)
print(merged_df.head())
print(merged_df.dtypes)
# Save the matched DataFrame to CSV
parent_folder = os.path.abspath(os.path.join(output_dirs[0], os.pardir))
output_filename = os.path.join(parent_folder, 'match_all.csv')
merged_df.to_csv(output_filename)
# Copy matched images to the new directory and rename them
for cam_idx, sync_output in enumerate(output_dirs):
if not osp.exists(sync_output):
os.makedirs(sync_output)
else:
shutil.rmtree(sync_output)
os.makedirs(sync_output)
for idx, row in merged_df.iterrows():
for cam_idx, vid in enumerate(input_dirs):
src_file = f"{vid}/{row[vid]}.png"
dst_file = os.path.join(output_dirs[cam_idx], f"{idx}.png")
shutil.copy(src_file, dst_file)
def rename2timestamp(target_dir, video_path):
# load frame timestamps csv, rename frames according to it
video_root, video_filename = os.path.split(video_path)
video_name, _ = os.path.splitext(video_filename)
video_date = re.sub(r"VID_((\d|_)*)", r"\1", video_name)
video_parent_dir = os.path.abspath(os.path.join(video_root, os.pardir))
with open(os.path.join(video_parent_dir, video_date + ".csv")) \
as frame_timestamps_file:
filename_timestamps = list(map(
lambda x: (x.strip('\n'), int(x)), frame_timestamps_file.readlines()
))
length = len(list(filter(
lambda x: os.path.splitext(x)[1] in ALLOWED_EXTENSIONS,
os.listdir(target_dir)
)))
length = min(length, len(filename_timestamps))
# frame number assertion
# assert len(filename_timestamps) == len(list(filter(
# lambda x: os.path.splitext(x)[1] in ALLOWED_EXTENSIONS,
# os.listdir(target_dir)
# ))), "Frame number in video %d and timestamp files %d did not match" % (l, len(filename_timestamps))
print(video_path, "=========================================================")
_, extension = os.path.splitext(os.listdir(target_dir)[0])
for i in range(length):
timestamp = filename_timestamps[i]
os.rename(
os.path.join(target_dir, "frame-%d.png" % (i + 1)),
os.path.join(target_dir, timestamp[0] + extension)
)
def extract_frames_from_videos(input_videos, extract_video_dirs):
for video_path, output_dir in zip(input_videos, extract_video_dirs):
if not osp.exists(output_dir):
os.makedirs(output_dir)
else:
shutil.rmtree(output_dir)
os.makedirs(output_dir)
# Construct and call the FFmpeg command
command = [
'ffmpeg', '-i', video_path, '-vsync', '0',
os.path.join(output_dir, 'frame-%d.png')
]
subprocess.run(command, check=True)
rename2timestamp(output_dir, video_path)
def sync_videos():
recsync_videos_root = "./capturesync/"
# input_videos = ["20240716/S22/VID/VID_20240716_141453.mp4", "20240716/Fold3/VID/VID_20240716_141453.mp4",
# "20240716/S20/VID/VID_20240716_141453.mp4", "20240716/TabS7/VID/VID_20240716_141453.mp4"]
input_videos = ["20240715/S22/VID/VID_20240715_170336.mp4", "20240715/Note9/VID/VID_20240715_170334.mp4",
"20240715/S20/VID/VID_20240715_170336.mp4", "20240715/TabS7/VID/VID_20240715_170336.mp4"]
input_videos = [os.path.join(recsync_videos_root, video) for video in input_videos]
output_dir = "./GMC0715"
extract_dir_temp = "./GMC0715/extract_frames/"
devices_name = [os.path.normpath(video).split(os.sep)[-3] for video in input_videos]
extract_video_dirs = [os.path.join(extract_dir_temp, devicen) for devicen in devices_name]
extract_frames_from_videos(input_videos, extract_video_dirs)
sync_dirs = [os.path.join(output_dir, devicen) for devicen in devices_name]
sync_all_frames(extract_video_dirs, sync_dirs)
# shutil.rmtree(extract_dir_temp)
visual_sync_output(sync_dirs, devices_name)
if __name__ == '__main__':
sync_videos()
calibration_syncvideos()