# -*- coding: utf-8 -*- """ Aplikasi Gradio untuk Analisis Komparatif Deteksi Helm Keselamatan ================================================================== Deskripsi: Aplikasi ini memungkinkan pengguna untuk mengunggah video dari lingkungan konstruksi dan membandingkan kinerja dua model AI (YOLOv5m dan YOLOv8m) dalam mendeteksi helm keselamatan secara real-time. """ # --- 1. IMPORT LIBRARY --- import gradio as gr from ultralyticsplus import YOLO, render_result import cv2 import numpy as np import os import tempfile import time import logging import subprocess # --- 2. SETUP & KONFIGURASI AWAL --- logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s') logger = logging.getLogger(__name__) # Set YOLO_CONFIG_DIR untuk mengatasi masalah izin os.environ["YOLO_CONFIG_DIR"] = "/tmp/Ultralytics" # --- 3. PEMUATAN MODEL AI --- try: logger.info("Memulai pemuatan model AI...") model_spesialis = YOLO('keremberke/yolov8m-hard-hat-detection') model_spesialis.overrides['conf'] = 0.25 model_spesialis.overrides['iou'] = 0.45 model_spesialis.overrides['agnostic_nms'] = False model_spesialis.overrides['max_det'] = 1000 logger.info("✅ Model Spesialis (YOLOv8m-HH) berhasil dimuat.") model_generalis = YOLO('keremberke/yolov5m-construction-safety') model_generalis.overrides['conf'] = 0.25 model_generalis.overrides['iou'] = 0.45 model_generalis.overrides['agnostic_nms'] = False model_generalis.overrides['max_det'] = 1000 logger.info("✅ Model Generalis (YOLOv5m-CS) berhasil dimuat.") models = { "YOLOv8m (Spesialis Helm)": model_spesialis, "YOLOv5m (Generalis Konstruksi)": model_generalis } logger.info("Semua model siap digunakan.") except Exception as e: logger.error(f"Gagal memuat model AI: {e}") raise RuntimeError(f"Tidak dapat memuat model AI. Error: {e}") # --- 4. FUNGSI KONVERSI VIDEO --- def convert_video_to_mp4(input_path): """Konversi video input ke MP4 jika format tidak didukung.""" temp_mp4_path = tempfile.NamedTemporaryFile(suffix=".mp4", delete=False).name ffmpeg_command = [ "ffmpeg", "-y", "-i", input_path, "-c:v", "libx264", "-preset", "veryfast", "-pix_fmt", "yuv420p", "-t", "30", temp_mp4_path ] result = subprocess.run(ffmpeg_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True) if result.returncode == 0 and os.path.exists(temp_mp4_path) and os.path.getsize(temp_mp4_path) > 0: logger.info("Video dikonversi ke MP4: %s", temp_mp4_path) if input_path != temp_mp4_path: os.remove(input_path) return temp_mp4_path else: logger.error("Konversi video gagal: %s", result.stderr) return None # --- 5. FUNGSI UTAMA PEMROSESAN VIDEO --- def process_video_and_analyze(video_path, selected_model_name, progress=gr.Progress(track_tqdm=True)): """Fungsi utama untuk memproses video, deteksi objek, dan analisis.""" if video_path is None: return None, "Status: Silakan unggah video terlebih dahulu.", "" try: logger.info(f"Memulai pemrosesan video: {video_path} menggunakan model: {selected_model_name}") start_time = time.time() # Pilih model model = models[selected_model_name] # Buka video dan coba konversi jika gagal cap = cv2.VideoCapture(video_path) if not cap.isOpened(): logger.warning("Format video tidak didukung, mencoba konversi ke MP4...") converted_path = convert_video_to_mp4(video_path) if converted_path: video_path = converted_path cap = cv2.VideoCapture(video_path) if not cap.isOpened(): logger.error("Gagal membuka video setelah konversi") return None, "Error: Video tidak dapat dibuka meskipun dikonversi.", "" # Dapatkan properti video fps = cap.get(cv2.CAP_PROP_FPS) or 15 total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) max_duration = 30 max_frames = int(fps * max_duration) if total_frames > max_frames: logger.warning("Video lebih dari 30 detik, dipotong ke %d frame", max_frames) total_frames = max_frames width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) target_width = 640 target_height = int(height * (target_width / width)) if width > 0 else 480 # Konfigurasi video output temp_output_path = tempfile.NamedTemporaryFile(suffix=".mp4", delete=False).name fourcc = cv2.VideoWriter_fourcc(*"mp4v") out = cv2.VideoWriter(temp_output_path, fourcc, fps, (target_width, target_height)) if not out.isOpened(): logger.error("Gagal membuka VideoWriter") cap.release() return None, "Error: Gagal membuat file video sementara.", "" # Variabel untuk analisis detection_count = 0 helm_detected_count = 0 frame_count = 0 progress(0, desc="Memulai Pemrosesan...") while cap.isOpened() and frame_count < max_frames: ret, frame = cap.read() if not ret: break frame_count += 1 # Update progress progress(frame_count / total_frames, desc=f"Memproses Frame {frame_count}/{total_frames}") # Resize frame frame_resized = cv2.resize(frame, (target_width, target_height)) # Proses frame (skip setiap frame ke-3 untuk performa) if frame_count % 3 != 0: out.write(frame_resized) continue # Prediksi results = model.predict(frame_resized) annotated_frame = render_result(model=model, image=frame_resized, result=results[0]) annotated_frame = np.array(annotated_frame) # Konversi PIL ke NumPy annotated_frame = cv2.cvtColor(annotated_frame, cv2.COLOR_RGB2BGR) # Konversi RGB ke BGR # Hitung deteksi detection_count += len(results[0].boxes) for box in results[0].boxes: class_id = int(box.cls) class_name = model.names[class_id].lower() if class_name in ['hardhat', 'helmet']: helm_detected_count += 1 out.write(annotated_frame) end_time = time.time() processing_time = end_time - start_time cap.release() out.release() # Verifikasi file output if not os.path.exists(temp_output_path) or os.path.getsize(temp_output_path) == 0: logger.error("File video sementara tidak valid atau kosong") return None, "Error: File video sementara tidak valid.", "" logger.info(f"Video berhasil diproses dalam {processing_time:.2f} detik.") # Analisis analysis_text = f""" ### Analisis Kinerja Model: {selected_model_name} - **Waktu Proses Total:** {processing_time:.2f} detik - **Total Frame Diproses:** {frame_count} - **Jumlah Deteksi Keseluruhan:** {detection_count} objek - **Jumlah Deteksi Helm:** {helm_detected_count} objek **Catatan:** - **Model Spesialis (YOLOv8m):** Fokus pada helm, akurasi tinggi untuk 'Hardhat'/'Helmet'. - **Model Generalis (YOLOv5m):** Deteksi berbagai objek, akurasi helm mungkin lebih rendah. """ return temp_output_path, f"Status: Video berhasil diproses! ({processing_time:.2f} detik)", analysis_text except Exception as e: logger.error(f"Terjadi error saat memproses video: {e}", exc_info=True) return None, f"Error: Terjadi kesalahan - {e}", "" finally: if 'cap' in locals() and cap.isOpened(): cap.release() if 'out' in locals() and out.isOpened(): out.release() # --- 6. PEMBUATAN ANTARMUKA PENGGUNA (GRADIO UI) --- with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="sky")) as iface: gr.Markdown( """ # 🛡ī¸ Analisis Komparatif Deteksi Helm Keselamatan Aplikasi ini membandingkan kinerja dua model AI dalam mendeteksi helm keselamatan pada video konstruksi. """ ) with gr.Row(): with gr.Column(scale=1): gr.Markdown("### **Langkah 1: Konfigurasi & Unggah**") selected_model_ui = gr.Radio( choices=list(models.keys()), label="Pilih Model AI", value="YOLOv8m (Spesialis Helm)" ) video_input = gr.Video( label="Unggah Video" ) detect_btn = gr.Button("🚀 Mulai Deteksi", variant="primary") gr.Markdown("### **Cara Penggunaan**") gr.Markdown( """ 1. Pilih model AI. 2. Unggah video atau gunakan contoh. 3. Klik "Mulai Deteksi". 4. Lihat hasil dan analisis. """ ) with gr.Column(scale=1): gr.Markdown("### **Langkah 2: Hasil Deteksi**") video_output = gr.Video(label="Video Hasil Deteksi") status_text = gr.Textbox(label="Status Proses", interactive=False) gr.Markdown("### **Ringkasan Analisis**") analysis_output = gr.Markdown(label="Analisis Kinerja") # Contoh video gr.Examples( examples=[ ["video.mp4", "YOLOv8m (Spesialis Helm)"] ], inputs=[video_input, selected_model_ui], outputs=[video_output, status_text, analysis_output], fn=process_video_and_analyze, cache_examples=False, label="Contoh Video" ) # Informasi dan kredit with gr.Accordion("ℹī¸ Informasi & Kredit", open=False): gr.Markdown( """ ### **Identitas Prototipe** - **Pengembang:** Faisal Fahmi Yuliawan - **Tujuan:** Membandingkan deteksi helm dengan model AI. ### **Kredit** - **Model AI:** keremberke/yolov8m-hard-hat-detection, keremberke/yolov5m-construction-safety - **Teknologi:** Ultralytics YOLO, Gradio, OpenCV, FFmpeg """ ) detect_btn.click( fn=process_video_and_analyze, inputs=[video_input, selected_model_ui], outputs=[video_output, status_text, analysis_output] ) # --- 7. LUNCURKAN APLIKASI --- if __name__ == "__main__": iface.launch(debug=True)