app.py

import cv2
import numpy as np
import torch
from ultralytics import YOLO
from sort import Sort

# Load YOLOv12x model
MODEL_PATH = "yolov12x.pt"
model = YOLO(MODEL_PATH)

# COCO dataset class ID for truck
TRUCK_CLASS_ID = 7  # "truck"

# Initialize SORT tracker
tracker = Sort()

# Minimum confidence threshold for detection
CONFIDENCE_THRESHOLD = 0.5

# Distance threshold to avoid duplicate counts
DISTANCE_THRESHOLD = 50

# Dictionary to define keyword-based time intervals
TIME_INTERVALS = {
    "fixed": 2,    # Fixed camera video
    "moving": 5,    # Moving camera video
    "drone": 7,     # Drone footage (slower interval)
    "dashcam": 6,   # Dashcam video
    "highway": 7,   # Highway footage
    "city": 8,      # City road footage
}

def old_determine_time_interval(video_filename):
    """
    Determines the time interval based on keywords found in the video filename.
    Defaults to 7 seconds if no matching keyword is found.
    """
    for keyword, interval in TIME_INTERVALS.items():
        if keyword in video_filename:
            return interval
    return 5  # Default interval if no keyword matches

def determine_time_interval(video_filename):
    print(f"Checking filename: {video_filename}")  # Debugging
    for keyword, interval in TIME_INTERVALS.items():
        if keyword in video_filename:
            print(f"Matched keyword: {keyword} -> Interval: {interval}")  # Debugging
            return interval
    print("No keyword match, using default interval: 5")  # Debugging
    return 5  # Default interval if no keyword matches
    

def count_unique_trucks(video_path):
    cap = cv2.VideoCapture(video_path)
    if not cap.isOpened():
        return "Error: Unable to open video file."

    unique_truck_ids = set()
    truck_history = {}

    # Get FPS of the video
    fps = int(cap.get(cv2.CAP_PROP_FPS))

    # Extract filename from the path and convert to lowercase
    video_filename = os.path.basename(video_path).lower()

    # Determine the dynamic time interval based on filename keywords
    time_interval = determine_time_interval(video_filename)
    frame_skip = fps * time_interval  # Convert time interval to frame count

    frame_count = 0

    while True:
        ret, frame = cap.read()
        if not ret:
            break  # End of video

        frame_count += 1
        if frame_count % frame_skip != 0:
            continue  # Skip frames to process only every 5 seconds

        # Run YOLOv12x inference
        results = model(frame, verbose=False)

        detections = []
        for result in results:
            for box in result.boxes:
                class_id = int(box.cls.item())  # Get class ID
                confidence = float(box.conf.item())  # Get confidence score

                # Track only trucks
                if class_id == TRUCK_CLASS_ID and confidence > CONFIDENCE_THRESHOLD:
                    x1, y1, x2, y2 = map(int, box.xyxy[0])  # Get bounding box
                    detections.append([x1, y1, x2, y2, confidence])

        if len(detections) > 0:
            detections = np.array(detections)
            tracked_objects = tracker.update(detections)

            for obj in tracked_objects:
                truck_id = int(obj[4])  # Unique ID assigned by SORT
                x1, y1, x2, y2 = obj[:4]  # Get the bounding box coordinates

                truck_center = (x1 + x2) / 2, (y1 + y2) / 2  # Calculate the center of the truck

                # If truck is already in history, check the movement distance
                if truck_id in truck_history:
                    last_position = truck_history[truck_id]["position"]
                    distance = np.linalg.norm(np.array(truck_center) - np.array(last_position))

                    if distance > DISTANCE_THRESHOLD:
                        # If the truck moved significantly, count as new
                        unique_truck_ids.add(truck_id)

                else:
                    # If truck is not in history, add it
                    truck_history[truck_id] = {
                        "frame_count": frame_count,
                        "position": truck_center
                    }
                    unique_truck_ids.add(truck_id)

    cap.release()

    return {"Total Unique Trucks": len(unique_truck_ids)}

# Gradio UI function
def analyze_video(video_file):
    result = count_unique_trucks(video_file)
    return "\n".join([f"{key}: {value}" for key, value in result.items()])

# Define Gradio interface
import gradio as gr
iface = gr.Interface(
    fn=analyze_video,
    inputs=gr.Video(label="Upload Video"),
    outputs=gr.Textbox(label="Analysis Result"),
    title="YOLOv12x Unique Truck Counter",
    description="Upload a video to count unique trucks using YOLOv12x and SORT tracking."
)

# Launch the Gradio app
if __name__ == "__main__":
    iface.launch()