main.py

import pyttsx3
from threading import Thread
from queue import Queue
from ultralytics import YOLO
import cv2
import numpy as np
import time



def speak(q):
    engine = pyttsx3.init()
    engine.setProperty('rate', 235)
    engine.setProperty('volume', 1.0)

    while True:
        if not q.empty():
            label, distance, position = q.get()
            rounded_distance = round(distance * 2) / 2  # Round to integer or in steps of 0.5
            # IF IT SAYS A INT NUMBER, IT REMOVES THE .0 PART. IT SAYS DIRECTLY 2 INSTEAD OF 2.0.
            rounded_distance_str = str(int(rounded_distance)) if rounded_distance.is_integer() else str(rounded_distance)
            if label in class_avg_sizes:
                engine.say(f"{label} IS {rounded_distance_str} METERS ON {position}")
                engine.runAndWait()
            with queue.mutex:
                queue.queue.clear()
        else:
            time.sleep(0.1)  # To avoid busy waiting
            
queue = Queue()
t = Thread(target=speak, args=(queue,))
t.start()

def calculate_distance(box, frame_width, class_avg_sizes):
    object_width = box.xyxy[0, 2].item() - box.xyxy[0, 0].item()

    label = result.names[box.cls[0].item()]

    if label in class_avg_sizes:
        object_width *= class_avg_sizes[label]["width_ratio"]

    distance = (frame_width * 0.5) / np.tan(np.radians(70 / 2)) / (object_width + 1e-6)
    return round(distance, 2)


def get_position(frame_width, box):
    if box[0] < frame_width // 3:
        return "LEFT"
    elif box[0] < 2 * (frame_width // 3):
        return "FORWARD"
    else:
        return "RIGHT"


def blur_person(image, box):
    x, y, w, h = box.xyxy[0].cpu().numpy().astype(int)
    top_region = image[y:y+int(0.08 * h), x:x+w]
    blurred_top_region = cv2.GaussianBlur(top_region, (15, 15), 0)
    image[y:y+int(0.08 *h), x:x+w] = blurred_top_region
    return image


model = YOLO("gpModel.pt")
cap = cv2.VideoCapture("test_video.mp4")

class_avg_sizes = {
    "person": {"width_ratio": 2.5},
    "car": {"width_ratio": 0.37},
    "bicycle": {"width_ratio": 2.3},
    "motorcycle": {"width_ratio": 2.4},
    "bus": {"width_ratio": 0.3},
    "traffic light": {"width_ratio": 2.95},
    "stop sign": {"width_ratio": 2.55},
    "bench": {"width_ratio": 1.6},
    "cat": {"width_ratio": 1.9},
    "dog": {"width_ratio": 1.5},
}

pause = False
while cap.isOpened():
    if not pause:
        ret, frame = cap.read()
        results = model.predict(frame)
        result = results[0]
        nearest_object = None
        min_distance = float('inf')
        detected_objects = []

        for box in result.boxes:
            label = result.names[box.cls[0].item()]
            cords = [round(x) for x in box.xyxy[0].tolist()]
            colorGreen = (0, 255, 0)
            colorYellow = (0, 255, 255)
            colorBlue = (255, 0, 0)
            colorRed = (0, 0, 255)

            thickness = 2

            distance = calculate_distance(box, frame.shape[1], class_avg_sizes) #box, frame_width, class_avg_sizes

            if distance < min_distance:
                min_distance = distance
                nearest_object = (label, round(distance, 1), cords)
                detected_objects = [(label, round(distance, 1))]

             # THE CLOSEST RED OBJECT DOES NOT MATTER
             # HUMAN GREEN
             # CAR YELLOW
             # OTHERS ARE BLUE

            if label == "person":
                frame = blur_person(frame, box)
                cv2.rectangle(frame, (cords[0], cords[1]), (cords[2], cords[3]), colorGreen, thickness)
                cv2.putText(frame, f"{label} - {distance:.1f}m", (cords[0], cords[1] - 10), cv2.FONT_HERSHEY_SIMPLEX,
                            0.5, colorGreen, thickness)
            elif label == "car":
                cv2.rectangle(frame, (cords[0], cords[1]), (cords[2], cords[3]), colorYellow, thickness)
                cv2.putText(frame, f"{label} - {distance:.1f}m", (cords[0], cords[1] - 10), cv2.FONT_HERSHEY_SIMPLEX,
                            0.5, colorYellow, thickness)
            # else:
            elif label in class_avg_sizes:
                cv2.rectangle(frame, (cords[0], cords[1]), (cords[2], cords[3]), colorBlue, thickness)
                cv2.putText(frame, f"{label} - {distance:.1f}m", (cords[0], cords[1] - 10), cv2.FONT_HERSHEY_SIMPLEX,
                            0.5, colorBlue, thickness)

        #  en yakın
        if nearest_object:

            if nearest_object[0] in class_avg_sizes:  # coordinats
                cv2.rectangle(frame, (nearest_object[2][0], nearest_object[2][1]),(nearest_object[2][2], nearest_object[2][3]), (0, 0, 255), thickness)
                text = f"{nearest_object[0]} - {round(nearest_object[1], 1)}m"
                cv2.putText(frame, text, (nearest_object[2][0], nearest_object[2][1] - 10), cv2.FONT_HERSHEY_SIMPLEX,
                            0.5, colorRed, thickness)

            if nearest_object[1] <= 12.5:  # give audio feedback if the distance is smaller or larger than the specified value
                position = get_position(frame.shape[1], nearest_object[2]) #frame_width, box
                queue.put((nearest_object[0], nearest_object[1], position))  # label, distance, position

            detected_objects.clear()
    else:
        frame = cap.retrieve()[1]



    cv2.imshow('Audio World ', frame)

    key = cv2.waitKey(1) & 0xFF

    if key == ord('q'):
        break
    elif key == ord('p'):
        pause = not pause

cap.release()
cv2.destroyAllWindows()