Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,6 +1,5 @@
|
|
| 1 |
import os
|
| 2 |
import sys
|
| 3 |
-
import subprocess
|
| 4 |
import logging
|
| 5 |
import warnings
|
| 6 |
import cv2
|
|
@@ -16,10 +15,8 @@ from reportlab.lib.units import inch
|
|
| 16 |
from io import BytesIO
|
| 17 |
import base64
|
| 18 |
from retrying import retry
|
| 19 |
-
import uuid
|
| 20 |
-
from multiprocessing import Pool, cpu_count
|
| 21 |
-
from functools import partial
|
| 22 |
from collections import defaultdict
|
|
|
|
| 23 |
|
| 24 |
# ========================== # Configuration and Setup # ==========================
|
| 25 |
os.environ['YOLO_CONFIG_DIR'] = '/tmp/Ultralytics'
|
|
@@ -29,7 +26,7 @@ logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(
|
|
| 29 |
logger = logging.getLogger(__name__)
|
| 30 |
warnings.filterwarnings("ignore")
|
| 31 |
|
| 32 |
-
# ========================== #
|
| 33 |
class SafetyTracker:
|
| 34 |
def __init__(self, track_thresh=0.3, track_buffer=30, match_thresh=0.7, frame_rate=30):
|
| 35 |
self.track_thresh = track_thresh
|
|
@@ -37,13 +34,10 @@ class SafetyTracker:
|
|
| 37 |
self.match_thresh = match_thresh
|
| 38 |
self.frame_rate = frame_rate
|
| 39 |
self.next_id = 1
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
-
# Tracking stores
|
| 42 |
-
self.worker_tracks = {} # Active tracks
|
| 43 |
-
self.violation_history = defaultdict(dict) # {worker_id: {violation_type: last_detection_time}}
|
| 44 |
-
self.position_history = defaultdict(list) # {worker_id: [positions]}
|
| 45 |
-
|
| 46 |
-
# Violation cooldowns (seconds)
|
| 47 |
self.VIOLATION_COOLDOWNS = {
|
| 48 |
"no_helmet": 30.0,
|
| 49 |
"no_harness": 20.0,
|
|
@@ -52,8 +46,7 @@ class SafetyTracker:
|
|
| 52 |
"improper_tool_use": 15.0
|
| 53 |
}
|
| 54 |
|
| 55 |
-
def update(self, detections):
|
| 56 |
-
"""Update tracks with new detections using position-based matching"""
|
| 57 |
current_time = time.time()
|
| 58 |
new_violations = []
|
| 59 |
|
|
@@ -62,14 +55,11 @@ class SafetyTracker:
|
|
| 62 |
label = det['violation']
|
| 63 |
confidence = det['confidence']
|
| 64 |
|
| 65 |
-
# Match by position
|
| 66 |
worker_id = self._match_by_position(bbox, label)
|
| 67 |
-
|
| 68 |
if worker_id is None:
|
| 69 |
worker_id = self.next_id
|
| 70 |
self.next_id += 1
|
| 71 |
|
| 72 |
-
# Check if new violation
|
| 73 |
if self._is_new_violation(worker_id, label, current_time):
|
| 74 |
violation = {
|
| 75 |
'worker_id': worker_id,
|
|
@@ -81,62 +71,41 @@ class SafetyTracker:
|
|
| 81 |
new_violations.append(violation)
|
| 82 |
self.violation_history[worker_id][label] = current_time
|
| 83 |
|
| 84 |
-
# Update position history
|
| 85 |
-
x, y, w, h = bbox
|
| 86 |
-
self.position_history[worker_id].append((x, y))
|
| 87 |
-
|
| 88 |
-
# Update active tracks
|
| 89 |
self.worker_tracks[worker_id] = {
|
| 90 |
'bbox': bbox,
|
| 91 |
'last_seen': current_time,
|
| 92 |
'label': label
|
| 93 |
}
|
|
|
|
| 94 |
|
| 95 |
-
# Cleanup old tracks
|
| 96 |
self._cleanup_tracks(current_time)
|
| 97 |
-
|
| 98 |
return new_violations
|
| 99 |
|
| 100 |
def _match_by_position(self, bbox, label):
|
| 101 |
-
|
| 102 |
-
x, y, w, h = bbox
|
| 103 |
-
current_pos = (x, y)
|
| 104 |
-
|
| 105 |
for worker_id, positions in self.position_history.items():
|
| 106 |
-
# Only match if worker has had this violation type before
|
| 107 |
-
if label not in self.violation_history[worker_id]:
|
| 108 |
-
continue
|
| 109 |
-
|
| 110 |
-
# Check distance to historical positions
|
| 111 |
for pos in positions[-5:]: # Check last 5 positions
|
| 112 |
-
|
| 113 |
-
if distance < 100: # Within 100 pixels
|
| 114 |
return worker_id
|
| 115 |
return None
|
| 116 |
|
| 117 |
def _is_new_violation(self, worker_id, label, current_time):
|
| 118 |
-
"""Check if violation is new based on cooldown"""
|
| 119 |
if label not in self.violation_history[worker_id]:
|
| 120 |
return True
|
| 121 |
-
|
| 122 |
-
last_time = self.violation_history[worker_id][label]
|
| 123 |
-
cooldown = self.VIOLATION_COOLDOWNS.get(label, 10.0)
|
| 124 |
-
return (current_time - last_time) > cooldown
|
| 125 |
|
| 126 |
def _cleanup_tracks(self, current_time):
|
| 127 |
-
"""Remove inactive tracks"""
|
| 128 |
inactive_ids = [
|
| 129 |
-
|
| 130 |
if (current_time - track['last_seen']) > (self.track_buffer / self.frame_rate)
|
| 131 |
]
|
| 132 |
-
for
|
| 133 |
-
self.worker_tracks.pop(
|
| 134 |
-
self.position_history.pop(
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
self.violation_history.pop(id, None)
|
| 138 |
|
| 139 |
-
# ========================== #
|
| 140 |
CONFIG = {
|
| 141 |
"MODEL_PATH": "yolov8_safety.pt",
|
| 142 |
"FALLBACK_MODEL": "yolov8n.pt",
|
|
@@ -156,10 +125,10 @@ CONFIG = {
|
|
| 156 |
"improper_tool_use": (255, 255, 0)
|
| 157 |
},
|
| 158 |
"DISPLAY_NAMES": {
|
| 159 |
-
"no_helmet": "No Helmet
|
| 160 |
-
"no_harness": "No Harness
|
| 161 |
"unsafe_posture": "Unsafe Posture",
|
| 162 |
-
"unsafe_zone": "Unsafe Zone
|
| 163 |
"improper_tool_use": "Improper Tool Use"
|
| 164 |
},
|
| 165 |
"SF_CREDENTIALS": {
|
|
@@ -177,7 +146,7 @@ CONFIG = {
|
|
| 177 |
"improper_tool_use": 0.3
|
| 178 |
},
|
| 179 |
"FRAME_SKIP": 2,
|
| 180 |
-
"BATCH_SIZE": 8,
|
| 181 |
"SNAPSHOT_QUALITY": 90
|
| 182 |
}
|
| 183 |
|
|
@@ -188,9 +157,7 @@ logger.info(f"Using device: {device}")
|
|
| 188 |
def load_model():
|
| 189 |
try:
|
| 190 |
model_path = CONFIG["MODEL_PATH"] if os.path.exists(CONFIG["MODEL_PATH"]) else CONFIG["FALLBACK_MODEL"]
|
| 191 |
-
logger.info(f"Loading model: {model_path}")
|
| 192 |
if not os.path.exists(model_path):
|
| 193 |
-
logger.info("Downloading fallback model...")
|
| 194 |
torch.hub.download_url_to_file('https://github.com/ultralytics/assets/releases/download/v8.3.0/yolov8n.pt', model_path)
|
| 195 |
return YOLO(model_path).to(device)
|
| 196 |
except Exception as e:
|
|
@@ -199,35 +166,23 @@ def load_model():
|
|
| 199 |
|
| 200 |
model = load_model()
|
| 201 |
|
| 202 |
-
def preprocess_frame(frame):
|
| 203 |
-
"""Basic image enhancement"""
|
| 204 |
-
return cv2.convertScaleAbs(frame, alpha=1.2, beta=20)
|
| 205 |
-
|
| 206 |
def draw_detections(frame, detections):
|
| 207 |
-
|
| 208 |
-
result = frame.copy()
|
| 209 |
for det in detections:
|
| 210 |
x, y, w, h = det['bbox']
|
| 211 |
x1, y1 = int(x-w/2), int(y-h/2)
|
| 212 |
x2, y2 = int(x+w/2), int(y+h/2)
|
| 213 |
-
|
| 214 |
-
|
| 215 |
-
|
| 216 |
-
cv2.
|
| 217 |
-
|
| 218 |
-
(x1, y1-10), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (255,255,255), 2)
|
| 219 |
-
return result
|
| 220 |
|
| 221 |
def calculate_safety_score(violations):
|
| 222 |
-
|
| 223 |
-
"no_helmet": 25, "no_harness": 30,
|
| 224 |
-
"unsafe_posture": 20, "unsafe_zone": 35,
|
| 225 |
-
"improper_tool_use": 25
|
| 226 |
-
}
|
| 227 |
unique_violations = {v['violation'] for v in violations}
|
| 228 |
-
return max(0, 100 - sum(
|
| 229 |
|
| 230 |
-
# ========================== # Reporting Functions # ==========================
|
| 231 |
def generate_violation_pdf(violations, score):
|
| 232 |
try:
|
| 233 |
pdf_buffer = BytesIO()
|
|
@@ -237,16 +192,16 @@ def generate_violation_pdf(violations, score):
|
|
| 237 |
c.setFont("Helvetica-Bold", 16)
|
| 238 |
c.drawString(1*inch, 10*inch, "Safety Violation Report")
|
| 239 |
c.setFont("Helvetica", 12)
|
| 240 |
-
c.drawString(1*inch, 9.5*inch, f"
|
| 241 |
c.drawString(1*inch, 9*inch, f"Safety Score: {score}%")
|
| 242 |
|
| 243 |
-
# Violations
|
| 244 |
y = 8.5*inch
|
| 245 |
c.setFont("Helvetica-Bold", 14)
|
| 246 |
c.drawString(1*inch, y, "Violations Detected:")
|
| 247 |
y -= 0.3*inch
|
| 248 |
-
|
| 249 |
c.setFont("Helvetica", 10)
|
|
|
|
| 250 |
for v in violations:
|
| 251 |
text = f"Worker {v['worker_id']}: {CONFIG['DISPLAY_NAMES'][v['violation']]} at {v['timestamp']:.1f}s"
|
| 252 |
c.drawString(1.2*inch, y, text)
|
|
@@ -254,6 +209,7 @@ def generate_violation_pdf(violations, score):
|
|
| 254 |
if y < 1*inch:
|
| 255 |
c.showPage()
|
| 256 |
y = 10*inch
|
|
|
|
| 257 |
|
| 258 |
c.save()
|
| 259 |
pdf_buffer.seek(0)
|
|
@@ -272,85 +228,63 @@ def generate_violation_pdf(violations, score):
|
|
| 272 |
def connect_to_salesforce():
|
| 273 |
try:
|
| 274 |
sf = Salesforce(**CONFIG["SF_CREDENTIALS"])
|
| 275 |
-
|
| 276 |
return sf
|
| 277 |
except Exception as e:
|
| 278 |
logger.error(f"Salesforce connection failed: {e}")
|
| 279 |
raise
|
| 280 |
|
| 281 |
-
def
|
| 282 |
try:
|
| 283 |
-
sf = connect_to_salesforce()
|
| 284 |
-
|
| 285 |
# Create record
|
| 286 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 287 |
"Compliance_Score__c": score,
|
| 288 |
"Violations_Found__c": len(violations),
|
| 289 |
-
"Violations_Details__c":
|
| 290 |
-
|
| 291 |
-
for v in violations
|
| 292 |
-
),
|
| 293 |
-
"Status__c": "New"
|
| 294 |
-
}
|
| 295 |
|
| 296 |
-
|
| 297 |
-
|
| 298 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 299 |
|
| 300 |
-
|
| 301 |
-
pdf_url = ""
|
| 302 |
-
if pdf_buffer:
|
| 303 |
-
encoded = base64.b64encode(pdf_buffer.getvalue()).decode()
|
| 304 |
-
content_version = sf.ContentVersion.create({
|
| 305 |
-
"Title": f"Safety_Report_{record_id}",
|
| 306 |
-
"PathOnClient": "report.pdf",
|
| 307 |
-
"VersionData": encoded,
|
| 308 |
-
"FirstPublishLocationId": record_id
|
| 309 |
-
})
|
| 310 |
-
pdf_url = f"https://{sf.sf_instance}/sfc/servlet.shepherd/version/download/{content_version['id']}"
|
| 311 |
-
logger.info(f"PDF uploaded: {pdf_url}")
|
| 312 |
-
|
| 313 |
-
return record_id, pdf_url
|
| 314 |
except Exception as e:
|
| 315 |
logger.error(f"Salesforce upload failed: {e}")
|
| 316 |
-
return None,
|
| 317 |
|
| 318 |
# ========================== # Video Processing # ==========================
|
| 319 |
def process_video(video_data):
|
| 320 |
try:
|
| 321 |
os.makedirs(CONFIG["OUTPUT_DIR"], exist_ok=True)
|
| 322 |
-
|
| 323 |
-
# Save video
|
| 324 |
-
video_path = os.path.join(CONFIG["OUTPUT_DIR"], f"input_{int(time.time())}.mp4")
|
| 325 |
with open(video_path, "wb") as f:
|
| 326 |
f.write(video_data)
|
| 327 |
|
| 328 |
cap = cv2.VideoCapture(video_path)
|
| 329 |
-
if not cap.isOpened():
|
| 330 |
-
raise ValueError("Failed to open video")
|
| 331 |
-
|
| 332 |
fps = cap.get(cv2.CAP_PROP_FPS) or 30
|
| 333 |
-
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
| 334 |
tracker = SafetyTracker(frame_rate=fps)
|
| 335 |
snapshots = []
|
| 336 |
|
| 337 |
-
|
| 338 |
-
while True:
|
| 339 |
ret, frame = cap.read()
|
| 340 |
if not ret:
|
| 341 |
break
|
| 342 |
|
| 343 |
-
|
| 344 |
-
|
| 345 |
-
continue
|
| 346 |
-
|
| 347 |
-
# Process frame
|
| 348 |
-
frame = preprocess_frame(frame)
|
| 349 |
-
results = model(frame, verbose=False)[0]
|
| 350 |
|
| 351 |
-
# Get detections
|
| 352 |
detections = []
|
| 353 |
-
for box in results.boxes:
|
| 354 |
cls = int(box.cls)
|
| 355 |
label = CONFIG["VIOLATION_LABELS"].get(cls)
|
| 356 |
if label and box.conf > CONFIG["CONFIDENCE_THRESHOLDS"].get(label, 0.3):
|
|
@@ -360,72 +294,49 @@ def process_video(video_data):
|
|
| 360 |
'confidence': float(box.conf)
|
| 361 |
})
|
| 362 |
|
| 363 |
-
|
| 364 |
-
new_violations = tracker.update(detections)
|
| 365 |
|
| 366 |
-
# Capture snapshots for new violations
|
| 367 |
for violation in new_violations:
|
| 368 |
snapshot = draw_detections(frame.copy(), [violation])
|
| 369 |
timestamp = time.strftime("%Y%m%d_%H%M%S")
|
| 370 |
-
img_path = os.path.join(
|
| 371 |
-
|
| 372 |
-
|
| 373 |
-
)
|
| 374 |
-
cv2.imwrite(img_path, snapshot, [cv2.IMWRITE_JPEG_QUALITY, CONFIG["SNAPSHOT_QUALITY"]])
|
| 375 |
snapshots.append({
|
| 376 |
'path': img_path,
|
| 377 |
'url': f"{CONFIG['PUBLIC_URL_BASE']}{os.path.basename(img_path)}",
|
| 378 |
'violation': violation
|
| 379 |
})
|
| 380 |
-
|
| 381 |
-
# Update progress
|
| 382 |
-
if frame_count % 10 == 0:
|
| 383 |
-
progress = min(100, frame_count / total_frames * 100)
|
| 384 |
-
yield f"Processing... {progress:.1f}%", "", "", "", ""
|
| 385 |
|
| 386 |
-
|
| 387 |
-
|
| 388 |
cap.release()
|
| 389 |
os.remove(video_path)
|
| 390 |
|
| 391 |
-
|
| 392 |
-
|
| 393 |
-
{
|
| 394 |
-
'worker_id': worker_id,
|
| 395 |
-
'violation': violation_type,
|
| 396 |
-
'timestamp': detection_time
|
| 397 |
-
}
|
| 398 |
-
for worker_id, violations in tracker.violation_history.items()
|
| 399 |
-
for violation_type, detection_time in violations.items()
|
| 400 |
-
]
|
| 401 |
-
|
| 402 |
-
if not violations:
|
| 403 |
-
yield "No violations found", "Safety Score: 100%", "No snapshots", "N/A", "N/A"
|
| 404 |
return
|
| 405 |
|
| 406 |
-
score = calculate_safety_score(
|
| 407 |
-
pdf_path, pdf_url,
|
| 408 |
-
record_id, salesforce_url = push_report_to_salesforce(violations, score, pdf_path, pdf_buffer)
|
| 409 |
-
|
| 410 |
-
# Format output
|
| 411 |
-
violations_table = "| Violation | Worker ID | Time |\n|-----------|-----------|------|\n"
|
| 412 |
-
violations_table += "\n".join(
|
| 413 |
-
f"| {CONFIG['DISPLAY_NAMES'][v['violation']]} | {v['worker_id']} | {v['timestamp']:.1f}s |"
|
| 414 |
-
for v in violations
|
| 415 |
-
)
|
| 416 |
|
| 417 |
-
|
| 418 |
-
|
| 419 |
-
|
| 420 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 421 |
)
|
| 422 |
|
| 423 |
yield (
|
| 424 |
-
|
| 425 |
f"Safety Score: {score}%",
|
| 426 |
-
snapshots_md
|
| 427 |
f"Salesforce ID: {record_id or 'N/A'}",
|
| 428 |
-
|
| 429 |
)
|
| 430 |
|
| 431 |
except Exception as e:
|
|
@@ -435,30 +346,22 @@ def process_video(video_data):
|
|
| 435 |
yield f"Error: {str(e)}", "", "", "", ""
|
| 436 |
|
| 437 |
# ========================== # Gradio Interface # ==========================
|
| 438 |
-
def gradio_interface(
|
| 439 |
-
if not
|
| 440 |
return "Upload a video file", "", "", "", ""
|
| 441 |
-
|
| 442 |
-
|
| 443 |
-
|
| 444 |
-
video_data = f.read()
|
| 445 |
-
|
| 446 |
-
for output in process_video(video_data):
|
| 447 |
-
yield output
|
| 448 |
-
|
| 449 |
-
except Exception as e:
|
| 450 |
-
logger.error(f"Interface error: {e}")
|
| 451 |
-
yield f"Error: {str(e)}", "", "", "", ""
|
| 452 |
|
| 453 |
interface = gr.Interface(
|
| 454 |
fn=gradio_interface,
|
| 455 |
-
inputs=gr.Video(
|
| 456 |
outputs=[
|
| 457 |
-
gr.Markdown("
|
| 458 |
-
gr.Textbox(
|
| 459 |
-
gr.Markdown("
|
| 460 |
-
gr.Textbox(
|
| 461 |
-
gr.Textbox(
|
| 462 |
],
|
| 463 |
title="AI Safety Compliance Analyzer",
|
| 464 |
description="Detects PPE and safety violations in worksite videos"
|
|
|
|
| 1 |
import os
|
| 2 |
import sys
|
|
|
|
| 3 |
import logging
|
| 4 |
import warnings
|
| 5 |
import cv2
|
|
|
|
| 15 |
from io import BytesIO
|
| 16 |
import base64
|
| 17 |
from retrying import retry
|
|
|
|
|
|
|
|
|
|
| 18 |
from collections import defaultdict
|
| 19 |
+
from multiprocessing import cpu_count
|
| 20 |
|
| 21 |
# ========================== # Configuration and Setup # ==========================
|
| 22 |
os.environ['YOLO_CONFIG_DIR'] = '/tmp/Ultralytics'
|
|
|
|
| 26 |
logger = logging.getLogger(__name__)
|
| 27 |
warnings.filterwarnings("ignore")
|
| 28 |
|
| 29 |
+
# ========================== # Optimized Tracker Implementation (No Face Recognition) # ==========================
|
| 30 |
class SafetyTracker:
|
| 31 |
def __init__(self, track_thresh=0.3, track_buffer=30, match_thresh=0.7, frame_rate=30):
|
| 32 |
self.track_thresh = track_thresh
|
|
|
|
| 34 |
self.match_thresh = match_thresh
|
| 35 |
self.frame_rate = frame_rate
|
| 36 |
self.next_id = 1
|
| 37 |
+
self.worker_tracks = {}
|
| 38 |
+
self.violation_history = defaultdict(dict)
|
| 39 |
+
self.position_history = defaultdict(list)
|
| 40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
self.VIOLATION_COOLDOWNS = {
|
| 42 |
"no_helmet": 30.0,
|
| 43 |
"no_harness": 20.0,
|
|
|
|
| 46 |
"improper_tool_use": 15.0
|
| 47 |
}
|
| 48 |
|
| 49 |
+
def update(self, detections, frame):
|
|
|
|
| 50 |
current_time = time.time()
|
| 51 |
new_violations = []
|
| 52 |
|
|
|
|
| 55 |
label = det['violation']
|
| 56 |
confidence = det['confidence']
|
| 57 |
|
|
|
|
| 58 |
worker_id = self._match_by_position(bbox, label)
|
|
|
|
| 59 |
if worker_id is None:
|
| 60 |
worker_id = self.next_id
|
| 61 |
self.next_id += 1
|
| 62 |
|
|
|
|
| 63 |
if self._is_new_violation(worker_id, label, current_time):
|
| 64 |
violation = {
|
| 65 |
'worker_id': worker_id,
|
|
|
|
| 71 |
new_violations.append(violation)
|
| 72 |
self.violation_history[worker_id][label] = current_time
|
| 73 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
self.worker_tracks[worker_id] = {
|
| 75 |
'bbox': bbox,
|
| 76 |
'last_seen': current_time,
|
| 77 |
'label': label
|
| 78 |
}
|
| 79 |
+
self.position_history[worker_id].append((bbox[0], bbox[1]))
|
| 80 |
|
|
|
|
| 81 |
self._cleanup_tracks(current_time)
|
|
|
|
| 82 |
return new_violations
|
| 83 |
|
| 84 |
def _match_by_position(self, bbox, label):
|
| 85 |
+
x, y = bbox[0], bbox[1]
|
|
|
|
|
|
|
|
|
|
| 86 |
for worker_id, positions in self.position_history.items():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 87 |
for pos in positions[-5:]: # Check last 5 positions
|
| 88 |
+
if np.sqrt((x-pos[0])**2 + (y-pos[1])**2) < 100:
|
|
|
|
| 89 |
return worker_id
|
| 90 |
return None
|
| 91 |
|
| 92 |
def _is_new_violation(self, worker_id, label, current_time):
|
|
|
|
| 93 |
if label not in self.violation_history[worker_id]:
|
| 94 |
return True
|
| 95 |
+
return (current_time - self.violation_history[worker_id][label]) > self.VIOLATION_COOLDOWNS.get(label, 10.0)
|
|
|
|
|
|
|
|
|
|
| 96 |
|
| 97 |
def _cleanup_tracks(self, current_time):
|
|
|
|
| 98 |
inactive_ids = [
|
| 99 |
+
wid for wid, track in self.worker_tracks.items()
|
| 100 |
if (current_time - track['last_seen']) > (self.track_buffer / self.frame_rate)
|
| 101 |
]
|
| 102 |
+
for wid in inactive_ids:
|
| 103 |
+
self.worker_tracks.pop(wid, None)
|
| 104 |
+
self.position_history.pop(wid, None)
|
| 105 |
+
if (current_time - max(self.violation_history[wid].values(), default=0)) > 300:
|
| 106 |
+
self.violation_history.pop(wid, None)
|
|
|
|
| 107 |
|
| 108 |
+
# ========================== # Configuration # ==========================
|
| 109 |
CONFIG = {
|
| 110 |
"MODEL_PATH": "yolov8_safety.pt",
|
| 111 |
"FALLBACK_MODEL": "yolov8n.pt",
|
|
|
|
| 125 |
"improper_tool_use": (255, 255, 0)
|
| 126 |
},
|
| 127 |
"DISPLAY_NAMES": {
|
| 128 |
+
"no_helmet": "No Helmet",
|
| 129 |
+
"no_harness": "No Harness",
|
| 130 |
"unsafe_posture": "Unsafe Posture",
|
| 131 |
+
"unsafe_zone": "Unsafe Zone",
|
| 132 |
"improper_tool_use": "Improper Tool Use"
|
| 133 |
},
|
| 134 |
"SF_CREDENTIALS": {
|
|
|
|
| 146 |
"improper_tool_use": 0.3
|
| 147 |
},
|
| 148 |
"FRAME_SKIP": 2,
|
| 149 |
+
"BATCH_SIZE": 8,
|
| 150 |
"SNAPSHOT_QUALITY": 90
|
| 151 |
}
|
| 152 |
|
|
|
|
| 157 |
def load_model():
|
| 158 |
try:
|
| 159 |
model_path = CONFIG["MODEL_PATH"] if os.path.exists(CONFIG["MODEL_PATH"]) else CONFIG["FALLBACK_MODEL"]
|
|
|
|
| 160 |
if not os.path.exists(model_path):
|
|
|
|
| 161 |
torch.hub.download_url_to_file('https://github.com/ultralytics/assets/releases/download/v8.3.0/yolov8n.pt', model_path)
|
| 162 |
return YOLO(model_path).to(device)
|
| 163 |
except Exception as e:
|
|
|
|
| 166 |
|
| 167 |
model = load_model()
|
| 168 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 169 |
def draw_detections(frame, detections):
|
| 170 |
+
annotated = frame.copy()
|
|
|
|
| 171 |
for det in detections:
|
| 172 |
x, y, w, h = det['bbox']
|
| 173 |
x1, y1 = int(x-w/2), int(y-h/2)
|
| 174 |
x2, y2 = int(x+w/2), int(y+h/2)
|
| 175 |
+
color = CONFIG["CLASS_COLORS"][det['violation']]
|
| 176 |
+
cv2.rectangle(annotated, (x1, y1), (x2, y2), color, 2)
|
| 177 |
+
label = f"{CONFIG['DISPLAY_NAMES'][det['violation']]} (Worker {det['worker_id']})"
|
| 178 |
+
cv2.putText(annotated, label, (x1, y1-10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255,255,255), 2)
|
| 179 |
+
return annotated
|
|
|
|
|
|
|
| 180 |
|
| 181 |
def calculate_safety_score(violations):
|
| 182 |
+
penalties = {"no_helmet":25, "no_harness":30, "unsafe_posture":20, "unsafe_zone":35, "improper_tool_use":25}
|
|
|
|
|
|
|
|
|
|
|
|
|
| 183 |
unique_violations = {v['violation'] for v in violations}
|
| 184 |
+
return max(0, 100 - sum(penalties.get(v,0) for v in unique_violations))
|
| 185 |
|
|
|
|
| 186 |
def generate_violation_pdf(violations, score):
|
| 187 |
try:
|
| 188 |
pdf_buffer = BytesIO()
|
|
|
|
| 192 |
c.setFont("Helvetica-Bold", 16)
|
| 193 |
c.drawString(1*inch, 10*inch, "Safety Violation Report")
|
| 194 |
c.setFont("Helvetica", 12)
|
| 195 |
+
c.drawString(1*inch, 9.5*inch, f"Date: {time.strftime('%Y-%m-%d %H:%M:%S')}")
|
| 196 |
c.drawString(1*inch, 9*inch, f"Safety Score: {score}%")
|
| 197 |
|
| 198 |
+
# Violations List
|
| 199 |
y = 8.5*inch
|
| 200 |
c.setFont("Helvetica-Bold", 14)
|
| 201 |
c.drawString(1*inch, y, "Violations Detected:")
|
| 202 |
y -= 0.3*inch
|
|
|
|
| 203 |
c.setFont("Helvetica", 10)
|
| 204 |
+
|
| 205 |
for v in violations:
|
| 206 |
text = f"Worker {v['worker_id']}: {CONFIG['DISPLAY_NAMES'][v['violation']]} at {v['timestamp']:.1f}s"
|
| 207 |
c.drawString(1.2*inch, y, text)
|
|
|
|
| 209 |
if y < 1*inch:
|
| 210 |
c.showPage()
|
| 211 |
y = 10*inch
|
| 212 |
+
c.setFont("Helvetica", 10)
|
| 213 |
|
| 214 |
c.save()
|
| 215 |
pdf_buffer.seek(0)
|
|
|
|
| 228 |
def connect_to_salesforce():
|
| 229 |
try:
|
| 230 |
sf = Salesforce(**CONFIG["SF_CREDENTIALS"])
|
| 231 |
+
sf.describe()
|
| 232 |
return sf
|
| 233 |
except Exception as e:
|
| 234 |
logger.error(f"Salesforce connection failed: {e}")
|
| 235 |
raise
|
| 236 |
|
| 237 |
+
def upload_to_salesforce(sf, pdf_file, violations, score):
|
| 238 |
try:
|
|
|
|
|
|
|
| 239 |
# Create record
|
| 240 |
+
violations_text = "\n".join(
|
| 241 |
+
f"Worker {v['worker_id']}: {CONFIG['DISPLAY_NAMES'][v['violation']]} at {v['timestamp']:.1f}s"
|
| 242 |
+
for v in violations
|
| 243 |
+
)
|
| 244 |
+
|
| 245 |
+
record = sf.Safety_Video_Report__c.create({
|
| 246 |
"Compliance_Score__c": score,
|
| 247 |
"Violations_Found__c": len(violations),
|
| 248 |
+
"Violations_Details__c": violations_text
|
| 249 |
+
})
|
|
|
|
|
|
|
|
|
|
|
|
|
| 250 |
|
| 251 |
+
# Upload PDF
|
| 252 |
+
encoded = base64.b64encode(pdf_file.getvalue()).decode()
|
| 253 |
+
content = sf.ContentVersion.create({
|
| 254 |
+
"Title": f"Safety_Report_{int(time.time())}",
|
| 255 |
+
"PathOnClient": "report.pdf",
|
| 256 |
+
"VersionData": encoded,
|
| 257 |
+
"FirstPublishLocationId": record['id']
|
| 258 |
+
})
|
| 259 |
|
| 260 |
+
return record['id'], f"https://{sf.sf_instance}/sfc/servlet.shepherd/version/download/{content['id']}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 261 |
except Exception as e:
|
| 262 |
logger.error(f"Salesforce upload failed: {e}")
|
| 263 |
+
return None, None
|
| 264 |
|
| 265 |
# ========================== # Video Processing # ==========================
|
| 266 |
def process_video(video_data):
|
| 267 |
try:
|
| 268 |
os.makedirs(CONFIG["OUTPUT_DIR"], exist_ok=True)
|
| 269 |
+
video_path = os.path.join(CONFIG["OUTPUT_DIR"], f"temp_{int(time.time())}.mp4")
|
|
|
|
|
|
|
| 270 |
with open(video_path, "wb") as f:
|
| 271 |
f.write(video_data)
|
| 272 |
|
| 273 |
cap = cv2.VideoCapture(video_path)
|
|
|
|
|
|
|
|
|
|
| 274 |
fps = cap.get(cv2.CAP_PROP_FPS) or 30
|
|
|
|
| 275 |
tracker = SafetyTracker(frame_rate=fps)
|
| 276 |
snapshots = []
|
| 277 |
|
| 278 |
+
while cap.isOpened():
|
|
|
|
| 279 |
ret, frame = cap.read()
|
| 280 |
if not ret:
|
| 281 |
break
|
| 282 |
|
| 283 |
+
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
| 284 |
+
results = model(frame, verbose=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 285 |
|
|
|
|
| 286 |
detections = []
|
| 287 |
+
for box in results[0].boxes:
|
| 288 |
cls = int(box.cls)
|
| 289 |
label = CONFIG["VIOLATION_LABELS"].get(cls)
|
| 290 |
if label and box.conf > CONFIG["CONFIDENCE_THRESHOLDS"].get(label, 0.3):
|
|
|
|
| 294 |
'confidence': float(box.conf)
|
| 295 |
})
|
| 296 |
|
| 297 |
+
new_violations = tracker.update(detections, frame)
|
|
|
|
| 298 |
|
|
|
|
| 299 |
for violation in new_violations:
|
| 300 |
snapshot = draw_detections(frame.copy(), [violation])
|
| 301 |
timestamp = time.strftime("%Y%m%d_%H%M%S")
|
| 302 |
+
img_path = os.path.join(CONFIG["OUTPUT_DIR"], f"violation_{violation['worker_id']}_{timestamp}.jpg")
|
| 303 |
+
cv2.imwrite(img_path, cv2.cvtColor(snapshot, cv2.COLOR_RGB2BGR),
|
| 304 |
+
[cv2.IMWRITE_JPEG_QUALITY, CONFIG["SNAPSHOT_QUALITY"]])
|
|
|
|
|
|
|
| 305 |
snapshots.append({
|
| 306 |
'path': img_path,
|
| 307 |
'url': f"{CONFIG['PUBLIC_URL_BASE']}{os.path.basename(img_path)}",
|
| 308 |
'violation': violation
|
| 309 |
})
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 310 |
|
| 311 |
+
yield f"Processing frame {int(cap.get(cv2.CAP_PROP_POS_FRAMES))}...", "", "", "", ""
|
| 312 |
+
|
| 313 |
cap.release()
|
| 314 |
os.remove(video_path)
|
| 315 |
|
| 316 |
+
if not snapshots:
|
| 317 |
+
yield "No violations detected", "Safety Score: 100%", "No snapshots", "N/A", "N/A"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 318 |
return
|
| 319 |
|
| 320 |
+
score = calculate_safety_score([v['violation'] for v in snapshots])
|
| 321 |
+
pdf_path, pdf_url, pdf_file = generate_violation_pdf([v['violation'] for v in snapshots], score)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 322 |
|
| 323 |
+
if pdf_file:
|
| 324 |
+
record_id, sf_url = upload_to_salesforce(connect_to_salesforce(), pdf_file,
|
| 325 |
+
[v['violation'] for v in snapshots], score)
|
| 326 |
+
else:
|
| 327 |
+
record_id, sf_url = None, None
|
| 328 |
+
|
| 329 |
+
snapshots_md = "\n".join(
|
| 330 |
+
f"![{v['violation']['violation']}]({v['url']})"
|
| 331 |
+
for v in snapshots
|
| 332 |
)
|
| 333 |
|
| 334 |
yield (
|
| 335 |
+
"\n".join(f"- {v['violation']['violation']} (Worker {v['violation']['worker_id']})" for v in snapshots),
|
| 336 |
f"Safety Score: {score}%",
|
| 337 |
+
snapshots_md,
|
| 338 |
f"Salesforce ID: {record_id or 'N/A'}",
|
| 339 |
+
sf_url or pdf_url or "N/A"
|
| 340 |
)
|
| 341 |
|
| 342 |
except Exception as e:
|
|
|
|
| 346 |
yield f"Error: {str(e)}", "", "", "", ""
|
| 347 |
|
| 348 |
# ========================== # Gradio Interface # ==========================
|
| 349 |
+
def gradio_interface(video):
|
| 350 |
+
if not video:
|
| 351 |
return "Upload a video file", "", "", "", ""
|
| 352 |
+
|
| 353 |
+
for update in process_video(open(video, "rb").read()):
|
| 354 |
+
yield update
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 355 |
|
| 356 |
interface = gr.Interface(
|
| 357 |
fn=gradio_interface,
|
| 358 |
+
inputs=gr.Video(),
|
| 359 |
outputs=[
|
| 360 |
+
gr.Markdown("Detected Violations"),
|
| 361 |
+
gr.Textbox("Safety Score"),
|
| 362 |
+
gr.Markdown("Evidence Snapshots"),
|
| 363 |
+
gr.Textbox("Salesforce Record"),
|
| 364 |
+
gr.Textbox("Report URL")
|
| 365 |
],
|
| 366 |
title="AI Safety Compliance Analyzer",
|
| 367 |
description="Detects PPE and safety violations in worksite videos"
|