File size: 15,136 Bytes
07bcae5 b9aee76 07bcae5 1329575 07bcae5 1329575 07bcae5 1329575 07bcae5 10f371d 07bcae5 b9aee76 07bcae5 90c11db 07bcae5 af91dd4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 |
import os
import cv2
import numpy as np
from moviepy.editor import VideoFileClip
import tempfile
import time
from PIL import Image, ImageDraw, ImageFont
import math
from face_analysis import get_face_embedding, cluster_faces, organize_faces_by_person
from pose_analysis import calculate_posture_score, draw_pose_landmarks
from anomaly_detection import anomaly_detection
from visualization import plot_mse, plot_mse_histogram, plot_mse_heatmap
from utils import frame_to_timecode
import pandas as pd
from facenet_pytorch import MTCNN
import torch
import mediapipe as mp
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
mtcnn = MTCNN(keep_all=False, device=device, thresholds=[0.9, 0.9, 0.9], min_face_size=50)
mp_face_mesh = mp.solutions.face_mesh
face_mesh = mp_face_mesh.FaceMesh(static_image_mode=False, max_num_faces=1, min_detection_confidence=0.5)
mp_pose = mp.solutions.pose
pose = mp_pose.Pose(static_image_mode=False, min_detection_confidence=0.6, min_tracking_confidence=0.6)
def extract_frames(video_path, output_folder, desired_fps, progress_callback=None):
os.makedirs(output_folder, exist_ok=True)
clip = VideoFileClip(video_path)
original_fps = clip.fps
duration = clip.duration
total_frames = int(duration * original_fps)
step = max(1, original_fps / desired_fps)
total_frames_to_extract = int(total_frames / step)
frame_count = 0
for t in np.arange(0, duration, step / original_fps):
frame = clip.get_frame(t)
cv2.imwrite(os.path.join(output_folder, f"frame_{frame_count:04d}.jpg"), cv2.cvtColor(frame, cv2.COLOR_RGB2BGR))
frame_count += 1
if progress_callback:
progress = min(100, (frame_count / total_frames_to_extract) * 100)
progress_callback(progress, f"Extracting frame")
if frame_count >= total_frames_to_extract:
break
clip.close()
return frame_count, original_fps
def process_frames(frames_folder, aligned_faces_folder, frame_count, progress):
embeddings_by_frame = {}
posture_scores_by_frame = {}
posture_landmarks_by_frame = {}
aligned_face_paths = []
frame_files = sorted([f for f in os.listdir(frames_folder) if f.endswith('.jpg')])
for i, frame_file in enumerate(frame_files):
frame_num = int(frame_file.split('_')[1].split('.')[0])
frame_path = os.path.join(frames_folder, frame_file)
frame = cv2.imread(frame_path)
if frame is not None:
posture_score, posture_landmarks = calculate_posture_score(frame)
posture_scores_by_frame[frame_num] = posture_score
posture_landmarks_by_frame[frame_num] = posture_landmarks
boxes, probs = mtcnn.detect(frame)
if boxes is not None and len(boxes) > 0 and probs[0] >= 0.99:
x1, y1, x2, y2 = [int(b) for b in boxes[0]]
face = frame[y1:y2, x1:x2]
if face.size > 0:
results = face_mesh.process(cv2.cvtColor(face, cv2.COLOR_BGR2RGB))
if results.multi_face_landmarks and is_frontal_face(results.multi_face_landmarks[0].landmark):
aligned_face = face
if aligned_face is not None:
aligned_face_resized = cv2.resize(aligned_face, (160, 160))
output_path = os.path.join(aligned_faces_folder, f"frame_{frame_num}_face.jpg")
cv2.imwrite(output_path, aligned_face_resized)
aligned_face_paths.append(output_path)
embedding = get_face_embedding(aligned_face_resized)
embeddings_by_frame[frame_num] = embedding
progress((i + 1) / len(frame_files), f"Processing frame {i + 1} of {len(frame_files)}")
return embeddings_by_frame, posture_scores_by_frame, posture_landmarks_by_frame, aligned_face_paths
def process_video(video_path, anomaly_threshold, desired_fps, progress=None):
start_time = time.time()
output_folder = "output"
os.makedirs(output_folder, exist_ok=True)
GRAPH_COLORS = {
'facial_embeddings': 'navy',
'body_posture': 'purple'
}
with tempfile.TemporaryDirectory() as temp_dir:
aligned_faces_folder = os.path.join(temp_dir, 'aligned_faces')
organized_faces_folder = os.path.join(temp_dir, 'organized_faces')
os.makedirs(aligned_faces_folder, exist_ok=True)
os.makedirs(organized_faces_folder, exist_ok=True)
clip = VideoFileClip(video_path)
video_duration = clip.duration
clip.close()
progress(0, "Starting frame extraction")
frames_folder = os.path.join(temp_dir, 'extracted_frames')
def extraction_progress(percent, message):
progress(percent / 100, f"Extracting frames")
frame_count, original_fps = extract_frames(video_path, frames_folder, desired_fps, extraction_progress)
progress(1, "Frame extraction complete")
progress(0.3, "Processing frames")
embeddings_by_frame, posture_scores_by_frame, posture_landmarks_by_frame, aligned_face_paths = process_frames(
frames_folder, aligned_faces_folder,
frame_count,
progress)
if not aligned_face_paths:
raise ValueError("No faces were extracted from the video.")
progress(0.6, "Clustering faces")
embeddings = [embedding for _, embedding in embeddings_by_frame.items()]
clusters = cluster_faces(embeddings)
num_clusters = len(set(clusters))
progress(0.7, "Organizing faces")
organize_faces_by_person(embeddings_by_frame, clusters, aligned_faces_folder, organized_faces_folder)
progress(0.8, "Saving person data")
df, largest_cluster = save_person_data_to_csv(embeddings_by_frame, clusters, desired_fps,
original_fps, temp_dir, video_duration)
df['Seconds'] = df['Timecode'].apply(
lambda x: sum(float(t) * 60 ** i for i, t in enumerate(reversed(x.split(':')))))
progress(0.85, "Getting face samples")
face_samples = get_all_face_samples(organized_faces_folder, output_folder, largest_cluster)
progress(0.9, "Performing anomaly detection")
embedding_columns = [col for col in df.columns if col.startswith('Raw_Embedding_')]
X_embeddings = df[embedding_columns].values
try:
X_posture = np.array([posture_scores_by_frame.get(frame, None) for frame in df['Frame']])
X_posture = X_posture[X_posture != None].reshape(-1, 1)
if len(X_posture) == 0:
raise ValueError("No valid posture data found")
mse_embeddings, mse_posture = anomaly_detection(X_embeddings, X_posture)
progress(0.95, "Generating plots")
mse_plot_embeddings, anomaly_frames_embeddings = plot_mse(df, mse_embeddings, "Facial Features",
color=GRAPH_COLORS['facial_embeddings'],
anomaly_threshold=anomaly_threshold)
mse_histogram_embeddings = plot_mse_histogram(mse_embeddings, "MSE Distribution: Facial Features",
anomaly_threshold, color=GRAPH_COLORS['facial_embeddings'])
mse_plot_posture, anomaly_frames_posture = plot_mse(df, mse_posture, "Body Posture",
color=GRAPH_COLORS['body_posture'],
anomaly_threshold=anomaly_threshold)
mse_histogram_posture = plot_mse_histogram(mse_posture, "MSE Distribution: Body Posture",
anomaly_threshold, color=GRAPH_COLORS['body_posture'])
mse_heatmap_posture = plot_mse_heatmap(mse_posture, "Body Posture MSE Heatmap", df)
mse_heatmap_embeddings = plot_mse_heatmap(mse_embeddings, "Facial Features MSE Heatmap", df)
except Exception as e:
print(f"Error details: {str(e)}")
import traceback
traceback.print_exc()
return (f"Error in video processing: {str(e)}",) + (None,) * 14
progress(1.0, "Preparing results")
results = f"Number of persons detected: {num_clusters}\n\n"
results += "Breakdown:\n"
for cluster_id in range(num_clusters):
face_count = len([c for c in clusters if c == cluster_id])
results += f"Person {cluster_id + 1}: {face_count} face frames\n"
end_time = time.time()
execution_time = end_time - start_time
def add_timecode_to_image(image, timecode):
img_pil = Image.fromarray(image)
draw = ImageDraw.Draw(img_pil)
font = ImageFont.load_default()
draw.text((10, 10), timecode, (255, 0, 0), font=font)
return np.array(img_pil)
anomaly_faces_embeddings = []
for frame in anomaly_frames_embeddings:
face_path = os.path.join(aligned_faces_folder, f"frame_{frame}_face.jpg")
if os.path.exists(face_path):
face_img = cv2.imread(face_path)
if face_img is not None:
face_img = cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB)
timecode = df[df['Frame'] == frame]['Timecode'].iloc[0]
face_img_with_timecode = add_timecode_to_image(face_img, timecode)
anomaly_faces_embeddings.append(face_img_with_timecode)
anomaly_frames_posture_images = []
for frame in anomaly_frames_posture:
frame_path = os.path.join(frames_folder, f"frame_{frame:04d}.jpg")
if os.path.exists(frame_path):
frame_img = cv2.imread(frame_path)
if frame_img is not None:
frame_img = cv2.cvtColor(frame_img, cv2.COLOR_BGR2RGB)
pose_results = pose.process(frame_img)
if pose_results.pose_landmarks:
frame_img = draw_pose_landmarks(frame_img, pose_results.pose_landmarks)
timecode = df[df['Frame'] == frame]['Timecode'].iloc[0]
frame_img_with_timecode = add_timecode_to_image(frame_img, timecode)
anomaly_frames_posture_images.append(frame_img_with_timecode)
return (
execution_time,
results,
df,
mse_embeddings,
mse_posture,
mse_plot_embeddings,
mse_histogram_embeddings,
mse_plot_posture,
mse_histogram_posture,
mse_heatmap_embeddings,
mse_heatmap_posture,
face_samples["most_frequent"],
face_samples["others"],
anomaly_faces_embeddings,
anomaly_frames_posture_images,
aligned_faces_folder,
frames_folder
)
def is_frontal_face(landmarks, threshold=50):
nose_tip = landmarks[4]
left_chin = landmarks[234]
right_chin = landmarks[454]
nose_to_left = [left_chin.x - nose_tip.x, left_chin.y - nose_tip.y]
nose_to_right = [right_chin.x - nose_tip.x, right_chin.y - nose_tip.y]
dot_product = nose_to_left[0] * nose_to_right[0] + nose_to_left[1] * nose_to_right[1]
magnitude_left = math.sqrt(nose_to_left[0] ** 2 + nose_to_left[1] ** 2)
magnitude_right = math.sqrt(nose_to_right[0] ** 2 + nose_to_right[1] ** 2)
cos_angle = dot_product / (magnitude_left * magnitude_right)
angle = math.acos(cos_angle)
angle_degrees = math.degrees(angle)
return abs(180 - angle_degrees) < threshold
def save_person_data_to_csv(embeddings_by_frame, clusters, desired_fps, original_fps, output_folder, video_duration):
person_data = {}
for (frame_num, embedding), cluster in zip(embeddings_by_frame.items(), clusters):
if cluster not in person_data:
person_data[cluster] = []
person_data[cluster].append((frame_num, embedding))
largest_cluster = max(person_data, key=lambda k: len(person_data[k]))
data = person_data[largest_cluster]
data.sort(key=lambda x: x[0])
frames, embeddings = zip(*data)
embeddings_array = np.array(embeddings)
np.save(os.path.join(output_folder, 'face_embeddings.npy'), embeddings_array)
total_frames = max(frames)
timecodes = [frame_to_timecode(frame, total_frames, video_duration) for frame in frames]
df_data = {
'Frame': frames,
'Timecode': timecodes,
'Embedding_Index': range(len(embeddings))
}
for i in range(len(embeddings[0])):
df_data[f'Raw_Embedding_{i}'] = [embedding[i] for embedding in embeddings]
df = pd.DataFrame(df_data)
return df, largest_cluster
def get_all_face_samples(organized_faces_folder, output_folder, largest_cluster, max_samples=100):
face_samples = {"most_frequent": [], "others": []}
for cluster_folder in sorted(os.listdir(organized_faces_folder)):
if cluster_folder.startswith("person_"):
person_folder = os.path.join(organized_faces_folder, cluster_folder)
face_files = sorted([f for f in os.listdir(person_folder) if f.endswith('.jpg')])
if face_files:
cluster_id = int(cluster_folder.split('_')[1])
if cluster_id == largest_cluster:
for i, sample in enumerate(face_files[:max_samples]):
face_path = os.path.join(person_folder, sample)
output_path = os.path.join(output_folder, f"face_sample_most_frequent_{i:04d}.jpg")
face_img = cv2.imread(face_path)
if face_img is not None:
small_face = cv2.resize(face_img, (160, 160))
cv2.imwrite(output_path, small_face)
face_samples["most_frequent"].append(output_path)
if len(face_samples["most_frequent"]) >= max_samples:
break
else:
remaining_samples = max_samples - len(face_samples["others"])
if remaining_samples > 0:
for i, sample in enumerate(face_files[:remaining_samples]):
face_path = os.path.join(person_folder, sample)
output_path = os.path.join(output_folder, f"face_sample_other_{cluster_id:02d}_{i:04d}.jpg")
face_img = cv2.imread(face_path)
if face_img is not None:
small_face = cv2.resize(face_img, (160, 160))
cv2.imwrite(output_path, small_face)
face_samples["others"].append(output_path)
if len(face_samples["others"]) >= max_samples:
break
return face_samples |