SourceCode.py

import cv2
import numpy as np
from ultralytics import YOLO
 #working effectively
 
"""
 understanding the prob:
 build a collision warnin system that 
 detectcs vehicles such as car, buses, trucks and pedestrain/bikes
 warns the driver if any objet gets close
 avoivds false postives

 object detection: identifying what is in the frame
 proximity estimation: determing how close objects are
 false postives : ensuiring warnings only trigger for genuine threats 

 a car occupying 30 % of the frame height is likely within 5 meters
 a pedestrain at 15% might be 3 meters away.
 """
# Initialize
model = YOLO("yolov8n.pt") #yolov8 is pre-trained model to detect objects, openCV handle video input/output and drawing warningas
#cap = cv2.VideoCapture("speed4.mp4")
#cap = cv2.VideoCapture("camVideo.mp4")
cap = cv2.VideoCapture("speed2.mp4")


# Constants
TARGET_WIDTH, TARGET_HEIGHT = 1280, 720  #to make the frame size constant, it gives fixed resolution 
CROP_RATIO = 0.85  # Crop out bonnet (lower 15%) #we will ignore bottom 15% car hood 

# Proximity thresholds
MIN_DANGER_HEIGHT_LARGE = 0.25 * TARGET_HEIGHT  # For cars/buses/trucks
#if the vehicle's height is > 25% of the frame height-> its very close
MIN_DANGER_HEIGHT_SMALL = 0.15 * TARGET_HEIGHT  # For bikes/pedestrians
#pedestrain/bikes are smaller, so the height threshold is lower that is 0.15 
Y_LOW_THRESHOLD = 0.65 * TARGET_HEIGHT  # Objects in lower 35% of frame
#the object must be lower in the 35% of the frame to be considered a risk
CONFIDENCE_THRESHOLD = 0.7  # Reduce false positives

"""the above is object height threshold
closer object appear larger in the frame
closer objects also appear lower in the frame
(when looked from windhield dashcam)
therefore we will consider if a car is taller than 25% of the frame-> car/buses/trucks close
if a pedestrain is taller tahn 15% of the frame-> pedestrain/bicycle
y_low_threshold : this will check vertical postion, object below this lines are prioritized and are consider closer to car 
"""
# Class-specific warnings (generalized for pedestrians/bikes)
CLASS_WARNINGS = {
    0: "WARNING: PEDESTRIAN/BICYCLE!",  # Person
    2: "WARNING: CAR CLOSE!",  # Car
    3: "WARNING: BIKE CLOSE!",  # Motorcycle (optional: replace with general warning if needed)
    5: "WARNING: BUS CLOSE!",  # Bus
    7: "WARNING: TRUCK CLOSE!"  # Truck
} 
"""
its difficult to generalize for pedestrain and bikes
YOLOv8 confuse may confuse bikes and pesdestrain
we will make sepearte logic for it """

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

    # Resize and crop bonnet
    frame = cv2.resize(frame, (TARGET_WIDTH, TARGET_HEIGHT)) #it is help in keeping the frame size consistent
    cropped_frame = frame[:int(TARGET_HEIGHT * CROP_RATIO), :] #cropped the frame implemented, we are just cropping the height so as to ignore the hood if it comes
    #this willl help reduce the false postives 

    # Detect objects (classes: 0=person, 2=car, 3=motorcycle, 5=bus, 7=truck)
    results = model(cropped_frame, classes=[0, 2, 3, 5, 7], conf=CONFIDENCE_THRESHOLD)
    """
    classes=[0,2,3,5,7]: detect only
    0:pedestrain
    2:cars
    3: motorcyles
    5: buses
    7: trucks
    conf=confidence=threshold =0.7 : this would ignire low-confidence detections and reduce noise
    """
    
    warning_text = None
    for result in results:
        boxes = result.boxes.xyxy.cpu().numpy()
        class_ids = result.boxes.cls.cpu().numpy().astype(int)
        for box, cls_id in zip(boxes, class_ids):
            x1, y1, x2, y2 = box #these are bounding boxes coordinates
            bbox_height = y2 - y1 #height of the bounding box in pixel
            #closer objects appears larger in the frame and will have taller bounding box
            y_bottom = y2 #vertical position
            #as object closer to car appear lower in the frame, while a faraway one appears higher.
            # bottom edge of the bounding  

            # Class-specific logic
            if cls_id in [2, 5, 7]:  # Cars, buses, trucks
                if bbox_height > MIN_DANGER_HEIGHT_LARGE and y_bottom > Y_LOW_THRESHOLD: #trigger warning (e.g. "Car close")
                    #if the vehicle is taller than 25% of the frame and its bottom edge is in the lower 35%-> then trigger warning
                    warning_text = CLASS_WARNINGS[cls_id]
                    break 
            elif cls_id in [0, 3]:  # Pedestrians or bikes
                if bbox_height > MIN_DANGER_HEIGHT_SMALL and y_bottom > Y_LOW_THRESHOLD: #this condition is for pedestrain/bikes
                    #if the pedestrain is taller than 15% of the frame and low in the frame-> triggers a genral warning 
                    #trigger warning 
                    warning_text = CLASS_WARNINGS[0]  # General warning for pedestrians/bikes
                    break
        """example scenarios
        distant car:
        bbox_height=10%, y_bottom=40% -> no warning (too small +too high)
        close pedestrain:
        bbox_height=20%, y_bottom=70%-> pedestrain meets both thresholds
        approaching truck:
        bbox_height=35%, y_bottom=80%->" truck close


        y_bottom represents the bottom edge of the bounding box in pixels, it helps us avoid false alrams 

        y_low_threshold=0.65*frame_height 65% of frame height
        if y_Bottom>y_low_threshold:  objecct is in the lower 35% of the frame-> close 
         
        """

    # Display warning
    if warning_text:
        font_scale = TARGET_WIDTH / 1000
        cv2.putText(frame, warning_text,
                   (int(TARGET_WIDTH*0.05), int(TARGET_HEIGHT*0.1)),
                   cv2.FONT_HERSHEY_SIMPLEX, font_scale*1.5,
                   (0, 0, 255), 3, cv2.LINE_AA) # this lines draws the warning text on the frame

    cv2.imshow("Collision Warning", frame)
    if cv2.waitKey(1) == 27:
        break

cap.release()
cv2.destroyAllWindows()