video_objects/object_detection/yolo5saveobjects.py
2025-04-02 08:47:55 -04:00

128 lines
4.7 KiB
Python
Executable file

#!/usr/bin/env python3
import cv2
import torch
import yt_dlp
import subprocess
import time
import os
import numpy as np
from datetime import datetime
from flask import Flask, Response
import warnings
# Warning! saves objects in objectlist.txt to file /detections/$object
warnings.simplefilter("ignore", category=FutureWarning) #ignores torch warning
app = Flask(__name__)
YOUTUBE_URL = "https://www.youtube.com/watch?v=t45_gP7I82I" # Stream URL
CONFIDENCE_THRESHOLD = 0.4 # Confidence threshold
MODEL = "yolov5s" # YOLO model version
# Load YOLOv5 model everytime, this keeps it updated, make force_reload = False if not.
print("🔄 Loading YOLOv5 model...")
model = torch.hub.load("ultralytics/yolov5", "custom", path=MODEL, force_reload=True)
print("✅ YOLOv5 loaded successfully!")
# Load object list from file
OBJECT_LIST_FILE = "objectlist.txt"
if os.path.exists(OBJECT_LIST_FILE):
with open(OBJECT_LIST_FILE, "r") as f:
OBJECT_NAMES = set(line.strip().lower() for line in f if line.strip())
print(f"📜 Loaded {len(OBJECT_NAMES)} objects from {OBJECT_LIST_FILE}")
else:
print(f"⚠️ {OBJECT_LIST_FILE} not found! No filtering will be applied.")
OBJECT_NAMES = set() # Empty set (detects all objects)
# Ensure directories exist
os.makedirs("detections", exist_ok=True)
def get_stream_url():
"""Fetch 720p YouTube stream URL using yt-dlp."""
ydl_opts = {'quiet': True, 'format': 'bestvideo[height=720]'}
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
info_dict = ydl.extract_info(YOUTUBE_URL, download=False)
return info_dict.get("url", None)
def save_cropped_object(frame, bbox, label):
"""Save cropped object with 50px padding in its respective folder."""
x1, y1, x2, y2 = bbox
h, w, _ = frame.shape
# Add 50px padding while ensuring within image bounds
x1, y1 = max(0, x1 - 50), max(0, y1 - 50)
x2, y2 = min(w, x2 + 50), min(h, y2 + 50)
cropped = frame[y1:y2, x1:x2] # Crop the bounding box
# Create folder for the object if it doesn't exist
save_dir = os.path.join("detections", label)
os.makedirs(save_dir, exist_ok=True)
# Save with timestamp
filename = f"{datetime.now().strftime('%Y%m%d_%H%M%S_%f')}.jpg"
cv2.imwrite(os.path.join(save_dir, filename), cropped)
print(f"📸 Saved: {os.path.join(save_dir, filename)}")
def generate_frames():
"""Capture video frames, apply object detection, and stream as MJPEG."""
stream_url = get_stream_url()
if not stream_url:
print("❌ Failed to fetch stream URL!")
return
print("🎥 Starting FFmpeg stream...")
ffmpeg_process = subprocess.Popen([
"ffmpeg", "-re", "-i", stream_url,
"-f", "rawvideo", "-pix_fmt", "bgr24", "pipe:1"
], stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, bufsize=10**8)
frame_width, frame_height = 1280, 720 # Expecting 720p stream
while True:
raw_frame = ffmpeg_process.stdout.read(frame_width * frame_height * 3) # Read raw BGR frame
if not raw_frame:
print("❌ No frame received!")
break
frame = np.frombuffer(raw_frame, np.uint8).reshape((frame_height, frame_width, 3)) # Convert to NumPy array
# Run YOLO object detection
results = model(frame)
detections = results.pandas().xyxy[0] # Convert detections to Pandas DataFrame
if detections.empty:
print("No objects detected.")
else:
print(f"✅ Detected {len(detections)} objects!")
for _, row in detections.iterrows():
label = row["name"].lower()
confidence = row["confidence"]
if confidence > CONFIDENCE_THRESHOLD and (not OBJECT_NAMES or label in OBJECT_NAMES):
x1, y1, x2, y2 = int(row["xmin"]), int(row["ymin"]), int(row["xmax"]), int(row["ymax"])
save_cropped_object(frame, (x1, y1, x2, y2), label)
# Draw bounding boxes
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2) # Green box
cv2.putText(frame, f"{label} ({confidence:.2f})", (x1, y1 - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
# Encode and present the frame as JPEG
_, buffer = cv2.imencode('.jpg', frame)
yield (b'--frame\r\n'
b'Content-Type: image/jpeg\r\n\r\n' + buffer.tobytes() + b'\r\n')
@app.route('/video')
def video_feed():
"""Stream processed video frames."""
return Response(generate_frames(), mimetype='multipart/x-mixed-replace; boundary=frame')
if __name__ == '__main__':
print("Running at http://localhost:5000/video")
app.run(host='0.0.0.0', port=5000, debug=True, threaded=True)