app.py

import cv2
import torch
from transformers import DetrImageProcessor, DetrForObjectDetection, TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image
from datetime import datetime

# Ensure all required libraries are installed
try:
    import timm  # Required by DETR
except ImportError:
    raise ImportError("The 'timm' library is required but not installed. Install it using 'pip install timm'.")

# Load the DETR model for object detection (license plate detection)
try:
    detr_processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
    detr_model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
except Exception as e:
    raise RuntimeError(f"Error initializing DETR model: {e}")

# Load the TrOCR model for OCR (license plate text recognition)
try:
    trocr_processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
    trocr_model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten")
except Exception as e:
    raise RuntimeError(f"Error initializing TrOCR model: {e}")

def detect_license_plate(frame):
    """
    Detect license plates in a video frame using DETR.
    """
    # Convert the frame to a PIL image
    pil_image = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
    
    # Preprocess the image for DETR
    inputs = detr_processor(images=pil_image, return_tensors="pt")
    outputs = detr_model(**inputs)
    
    # Get detected objects and filter for license plates
    logits = outputs.logits
    boxes = outputs.pred_boxes
    probas = logits.softmax(-1)[0, :, :-1]
    keep = probas.max(-1).values > 0.9  # Confidence threshold
    
    detected_boxes = []
    for box, score in zip(boxes[keep], probas[keep]):
        # Convert box coordinates to pixel values
        box = box.detach().cpu().numpy()
        detected_boxes.append(box)
    
    return detected_boxes

def recognize_text(plate_image):
    """
    Recognize text from a license plate image using TrOCR.
    """
    # Convert the license plate image to a PIL image
    pil_image = Image.fromarray(cv2.cvtColor(plate_image, cv2.COLOR_BGR2RGB))
    
    # Preprocess the image for TrOCR
    pixel_values = trocr_processor(images=pil_image, return_tensors="pt").pixel_values
    generated_ids = trocr_model.generate(pixel_values)
    text = trocr_processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
    
    return text.strip()

def process_video(video_path, frame_skip=5):
    """
    Process a video to detect license plates and log entry/exit times.
    """
    cap = cv2.VideoCapture(video_path)
    vehicle_data = {}
    frame_count = 0

    while cap.isOpened():
        ret, frame = cap.read()
        if not ret:
            break

        frame_count += 1
        if frame_count % frame_skip != 0:
            continue  # Skip frames to optimize processing time

        # Detect license plates
        detected_boxes = detect_license_plate(frame)
        
        for box in detected_boxes:
            x_min, y_min, x_max, y_max = map(int, box)
            license_plate_image = frame[y_min:y_max, x_min:x_max]
            
            # Recognize text from the license plate
            license_plate = recognize_text(license_plate_image)
            
            if license_plate:
                current_time = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
                
                if license_plate not in vehicle_data:
                    # Vehicle entering
                    vehicle_data[license_plate] = {'entry_time': current_time, 'exit_time': None}
                    print(f"Vehicle {license_plate} entered at {current_time}")
                else:
                    # Update exit time
                    vehicle_data[license_plate]['exit_time'] = current_time
                
                # Draw bounding box and license plate text
                cv2.rectangle(frame, (x_min, y_min), (x_max, y_max), (0, 255, 0), 2)
                cv2.putText(frame, license_plate, (x_min, y_min-10), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 255, 0), 2)
        
        # Display the frame (optional, can be removed for headless environments)
        cv2.imshow('Vehicle Detection', frame)
        
        # Break on 'q' key press
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break

    cap.release()
    cv2.destroyAllWindows()
    
    # Print vehicle data
    print("\nVehicle Data:")
    for plate, times in vehicle_data.items():
        print(f"License Plate: {plate}, Entry Time: {times['entry_time']}, Exit Time: {times['exit_time']}")

if __name__ == "__main__":
    # Replace 'road_video.mp4' with the path to your video file or use 0 for webcam
    process_video("road_video.mp4", frame_skip=5)