# Open the video file
video_path = 'industrial_video.mp4'
cap = cv2.VideoCapture(video_path)

# Loop through the video frames
while cap.isOpened():
    # Read a frame from the video
    success, frame = cap.read()

    if success:
        # Run YOLOv8 inference on the frame
        results = model(frame)

        # A flag to check if fire was detected in the current frame
        fire_detected_in_frame = False

        # Visualize the results on the frame
        annotated_frame = results[0].plot()

        # Process detection results
        for r in results:
            for box in r.boxes:
                # Check if the detected class is 'fire'
                # model.names[0] should correspond to 'fire' in your custom model
                if model.names[int(box.cls[0])] == 'fire' and box.conf[0] > 0.5:
                    fire_detected_in_frame = True
                    break
        
        # If fire is detected and alarm is not already on, trigger alarm
        if fire_detected_in_frame and not alarm_on:
            alarm_on = True
            # Run the alarm sound in a background thread to not block video feed
            alarm_thread = threading.Thread(target=play_alarm)
            alarm_thread.start()

        # Display the annotated frame
        cv2.imshow("YOLOv8 Fire Detection", annotated_frame)

        # Break the loop if 'q' is pressed
        if cv2.waitKey(1) & 0xFF == ord("q"):
            break
    else:
        # Break the loop if the end of the video is reached
        break

# Release the video capture object and close the display window
cap.release()
cv2.destroyAllWindows()

# Hashtags: #RealTimeDetection #VideoProcessing #OpenCV

---

#Step 4: Results and Discussion

After running the script, you will see a window playing the video. When the model detects an object it identifies as 'fire' with a confidence score above 50%, it will:
• Draw a colored box around the fire.
• Print "ALARM: Fire Detected!" to the console.
• Play the alarm.wav sound.

Discussion of Results:
Model Performance: The accuracy of this system depends entirely on the quality of your custom-trained model (fire_model.pt). A model trained on a diverse dataset of industrial fires (different lighting, angles, sizes) will perform best.
False Positives: The system might incorrectly identify orange/red lights, reflections, or welding sparks as fire. This is a common challenge. To fix this, you need to add more "negative" images (images of things that look like fire but aren't) to your training dataset.
Thresholding: The confidence threshold (box.conf[0] > 0.5) is a critical parameter. A lower value increases the chance of detecting real fires but also increases false alarms. A higher value reduces false alarms but might miss smaller or less obvious fires. You must tune this value based on your specific environment.
Real-World Implementation: For a real industrial facility, you would replace the video file with a live camera stream (cv2.VideoCapture(0) for a webcam) and integrate the alarm logic with a physical siren or a central monitoring system via an API or GPIO pins.

#ProjectComplete #AIforGood #IndustrialSafety

━━━━━━━━━━━━━━━
By: @DataScience4 ✨