import cv2 import torch import gradio as gr import numpy as np import os import json import logging import matplotlib.pyplot as plt from datetime import datetime from collections import Counter from typing import List, Dict, Any, Optional from ultralytics import YOLO import ultralytics import time # Set YOLO config directory os.environ["YOLO_CONFIG_DIR"] = "/tmp/Ultralytics" # Set up logging logging.basicConfig( filename="app.log", level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s" ) # Directories CAPTURED_FRAMES_DIR = "captured_frames" OUTPUT_DIR = "outputs" os.makedirs(CAPTURED_FRAMES_DIR, exist_ok=True) os.makedirs(OUTPUT_DIR, exist_ok=True) os.chmod(CAPTURED_FRAMES_DIR, 0o777) os.chmod(OUTPUT_DIR, 0o777) # Global variables log_entries: List[str] = [] detected_counts: List[int] = [] detected_issues: List[str] = [] gps_coordinates: List[List[float]] = [] last_metrics: Dict[str, Any] = {} frame_count: int = 0 SAVE_IMAGE_INTERVAL = 1 # Save every frame with detections # Debug: Check environment print(f"Torch version: {torch.__version__}") print(f"Gradio version: {gr.__version__}") print(f"Ultralytics version: {ultralytics.__version__}") print(f"CUDA available: {torch.cuda.is_available()}") # Load custom YOLO model device = "cuda" if torch.cuda.is_available() else "cpu" print(f"Using device: {device}") model = YOLO('./data/best.pt').to(device) if device == "cuda": model.half() # Use half-precision (FP16) print(f"Model classes: {model.names}") # Mock service functions def generate_map(gps_coords: List[List[float]], items: List[Dict[str, Any]]) -> str: map_path = "map_temp.png" plt.figure(figsize=(4, 4)) plt.scatter([x[1] for x in gps_coords], [x[0] for x in gps_coords], c='blue', label='GPS Points') plt.title("Issue Locations Map") plt.xlabel("Longitude") plt.ylabel("Latitude") plt.legend() plt.savefig(map_path) plt.close() return map_path def send_to_salesforce(data: Dict[str, Any]) -> None: pass # Minimal mock def update_metrics(detections: List[Dict[str, Any]]) -> Dict[str, Any]: counts = Counter([det["label"] for det in detections]) return { "items": [{"type": k, "count": v} for k, v in counts.items()], "total_detections": len(detections), "timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S") } def generate_line_chart() -> Optional[str]: if not detected_counts: return None plt.figure(figsize=(4, 2)) plt.plot(detected_counts[-50:], marker='o', color='#FF8C00') plt.title("Detections Over Time") plt.xlabel("Frame") plt.ylabel("Count") plt.grid(True) plt.tight_layout() chart_path = "chart_temp.png" plt.savefig(chart_path) plt.close() return chart_path def process_video(video, resize_width=320, resize_height=240, frame_skip=5): global frame_count, last_metrics, detected_counts, detected_issues, gps_coordinates, log_entries frame_count = 0 detected_counts.clear() detected_issues.clear() gps_coordinates.clear() log_entries.clear() last_metrics = {} if video is None: log_entries.append("Error: No video uploaded") logging.error("No video uploaded") return "processed_output.mp4", json.dumps({"error": "No video uploaded"}, indent=2), "\n".join(log_entries), [], None, None start_time = time.time() cap = cv2.VideoCapture(video) if not cap.isOpened(): log_entries.append("Error: Could not open video file") logging.error("Could not open video file") return "processed_output.mp4", json.dumps({"error": "Could not open video file"}, indent=2), "\n".join(log_entries), [], None, None frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) fps = cap.get(cv2.CAP_PROP_FPS) total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) expected_duration = total_frames / fps if fps > 0 else 0 log_entries.append(f"Input video: {frame_width}x{frame_height}, {fps} FPS, {total_frames} frames, {expected_duration:.2f} seconds") logging.info(f"Input video: {frame_width}x{frame_height}, {fps} FPS, {total_frames} frames, {expected_duration:.2f} seconds") print(f"Input video: {frame_width}x{frame_height}, {fps} FPS, {total_frames} frames, {expected_duration:.2f} seconds") out_width, out_height = resize_width, resize_height output_path = "processed_output.mp4" codecs = [('mp4v', '.mp4'), ('MJPG', '.avi'), ('XVID', '.avi')] out = None for codec, ext in codecs: fourcc = cv2.VideoWriter_fourcc(*codec) output_path = f"processed_output{ext}" out = cv2.VideoWriter(output_path, fourcc, fps, (out_width, out_height)) if out.isOpened(): log_entries.append(f"Using codec: {codec}, output: {output_path}") logging.info(f"Using codec: {codec}, output: {output_path}") break else: log_entries.append(f"Failed to initialize codec: {codec}") logging.warning(f"Failed to initialize codec: {codec}") if not out or not out.isOpened(): log_entries.append("Error: All codecs failed to initialize video writer") logging.error("All codecs failed to initialize video writer") cap.release() return "processed_output.mp4", json.dumps({"error": "All codecs failed"}, indent=2), "\n".join(log_entries), [], None, None processed_frames = 0 all_detections = [] frame_times = [] detection_frame_count = 0 output_frame_count = 0 last_annotated_frame = None while True: ret, frame = cap.read() if not ret: break frame_count += 1 if frame_count % frame_skip != 0: continue processed_frames += 1 frame Cheesecake = time.time() frame = cv2.resize(frame, (out_width, out_height)) results = model(frame, verbose=False, conf=0.5, iou=0.7) annotated_frame = results[0].plot() # Calculate timestamp for the current frame frame_timestamp = frame_count / fps if fps > 0 else 0 timestamp_str = f"{int(frame_timestamp // 60)}:{int(frame_timestamp % 60):02d}" frame_detections = [] for detection in results[0].boxes: cls = int(detection.cls) conf = float(detection.conf) box = detection.xyxy[0].cpu().numpy().astype(int).tolist() label = model.names[cls] if label != 'Crocodile': # Ignore irrelevant class frame_detections.append({"label": label, "box": box, "conf": conf}) # Log detection with timestamp log_message = f"Frame {frame_count} at {timestamp_str}: Detected {label} with confidence {conf:.2f}" log_entries.append(log_message) logging.info(log_message) if frame_detections: detection_frame_count += 1 if detection_frame_count % SAVE_IMAGE_INTERVAL == 0: captured_frame_path = os.path.join(CAPTURED_FRAMES_DIR, f"detected_{frame_count}.jpg") if not cv2.imwrite(captured_frame_path, annotated_frame): log_entries.append(f"Error: Failed to save {captured_frame_path}") logging.error(f"Failed to save {captured_frame_path}") else: detected_issues.append(captured_frame_path) if len(detected_issues) > 100: detected_issues.pop(0) # Write frame and duplicates out.write(annotated_frame) output_frame_count += 1 last_annotated_frame = annotated_frame if frame_skip > 1: for _ in range(frame_skip - 1): out.write(annotated_frame) output_frame_count += 1 detected_counts.append(len(frame_detections)) gps_coord = [17.385044 + (frame_count * 0.0001), 78.486671 + (frame_count * 0.0001)] gps_coordinates.append(gps_coord) for det in frame_detections: det["gps"] = gps_coord det["timestamp"] = timestamp_str # Add timestamp to detection data all_detections.extend(frame_detections) frame_time = (time.time() - frame_start) * 1000 frame_times.append(frame_time) detection_summary = { "timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"), "video_timestamp": timestamp_str, "frame": frame_count, "longitudinal": sum(1 for det in frame_detections if det["label"] == "Longitudinal"), "pothole": sum(1 for det in frame_detections if det["label"] == "Pothole"), "transverse": sum(1 for det in frame_detections if det["label"] == "Transverse"), "gps": gps_coord, "processing_time_ms": frame_time } log_entries.append(json.dumps(detection_summary, indent=2)) if len(log_entries) > 50: log_entries.pop(0) # Pad remaining frames while output_frame_count < total_frames and last_annotated_frame is not None: out.write(last_annotated_frame) output_frame_count += 1 last_metrics = update_metrics(all_detections) send_to_salesforce({ "detections": all_detections, "metrics": last_metrics, "timestamp": detection_summary["timestamp"] if all_detections else datetime.now().strftime("%Y-%m-%d %H:%M:%S"), "frame_count": frame_count, "gps_coordinates": gps_coordinates[-1] if gps_coordinates else [0, 0] }) cap.release() out.release() cap = cv2.VideoCapture(output_path) output_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) output_fps = cap.get(cv2.CAP_PROP_FPS) output_duration = output_frames / output_fps if output_fps > 0 else 0 cap.release() total_time = time.time() - start_time avg_frame_time = sum(frame_times) / len(frame_times) if frame_times else 0 log_entries.append(f"Output video: {output_frames} frames, {output_fps} FPS, {output_duration:.2f} seconds") log_entries.append(f"Total processing time: {total_time:.2f} seconds, Avg frame time: {avg_frame_time:.2f} ms, Detection frames: {detection_frame_count}, Output frames: {output_frame_count}") logging.info(f"Output video: {output_frames} frames, {output_fps} FPS, {output_duration:.2f} seconds") logging.info(f"Total processing time: {total_time:.2f} seconds, Avg frame time: {avg_frame_time:.2f} ms, Detection frames: {detection_frame_count}, Output frames: {output_frame_count}") print(f"Output video: {output_frames} frames, {output_fps} FPS, {output_duration:.2f} seconds") print(f"Total processing time: {total_time:.2f} seconds, Avg frame time: {avg_frame_time:.2f} ms, Detection frames: {detection_frame_count}, Output frames: {output_frame_count}") chart_path = generate_line_chart() map_path = generate_map(gps_coordinates[-5:], all_detections) return ( output_path, json.dumps(last_metrics, indent=2), "\n".join(log_entries[-10:]), detected_issues, chart_path, map_path ) # Gradio interface with gr.Blocks(theme=gr.themes.Soft(primary_hue="orange")) as iface: gr.Markdown("# Road Defect Detection Dashboard") with gr.Row(): with gr.Column(scale=3): video_input = gr.Video(label="Upload Video") width_slider = gr.Slider(320, 640, value=320, label="Output Width", step=1) height_slider = gr.Slider(240, 480, value=240, label="Output Height", step=1) skip_slider = gr.Slider(1, 10, value=5, label="Frame Skip", step=1) process_btn = gr.Button("Process Video", variant="primary") with gr.Column(scale=1): metrics_output = gr.Textbox(label="Detection Metrics", lines=5, interactive=False) with gr.Row(): video_output = gr.Video(label="Processed Video") issue_gallery = gr.Gallery(label="Detected Issues", columns=4, height="auto", object_fit="contain") with gr.Row(): chart_output = gr.Image(label="Detection Trend") map_output = gr.Image(label="Issue Locations Map") with gr.Row(): logs_output = gr.Textbox(label="Logs", lines=5, interactive=False) process_btn.click( process_video, inputs=[video_input, width_slider, height_slider, skip_slider], outputs=[video_output, metrics_output, logs_output, issue_gallery, chart_output, map_output] ) if __name__ == "__main__": iface.launch()