I am tried so many time to figure out the problem in detecting the face and also it’s not so smooth enough to like other tools out there.

So basically I am using python and Yolo in this project but I want the person who is talking and who the ROI (region of interest) is.

Here is the code:

`from ultralytics import YOLO

from ultralytics.engine.results import Results

from moviepy.editor import VideoFileClip, concatenate_videoclips

from moviepy.video.fx.crop import crop

Load the YOLOv8 model

model = YOLO(“yolov8n.pt”)

Load the input video

clip = VideoFileClip(“short_test.mp4”)

tacked_clips = []

for frame_no, frame in enumerate(clip.iter_frames()):

# Process the frame

results: list[Results] = model(frame)



# Get the bounding box of the main object

if results[0].boxes:

    objects = results[0].boxes

    main_obj = max(

        objects, key=lambda x: x.conf

    )  # Assuming the first detected object is the main one



    x1, y1, x2, y2 = [int(val) for val in main_obj.xyxy[0].tolist()]



    # Calculate the crop region based on the object's position and the target aspect ratio

    w, h = clip.size

    new_w = int(h * 9 / 16)

    new_h = h



    x_center = x2 - x1

    y_center = y2 - y1



    # Adjust x_center and y_center if they would cause the crop region to exceed the bounds

    if x_center + (new_w / 2) > w:

        x_center -= x_center + (new_w / 2) - w

    elif x_center - (new_w / 2) < 0:

        x_center += abs(x_center - (new_w / 2))



    if y_center + (new_h / 2) > h:

        y_center -= y_center + (new_h / 2) - h

    elif y_center - (new_h / 2) < 0:

        y_center += abs(y_center - (new_h / 2))



    # Create a subclip for the current frame

    start_time = frame_no / clip.fps

    end_time = (frame_no + 1) / clip.fps

    subclip = clip.subclip(start_time, end_time)



    # Apply cropping using MoviePy

    cropped_clip = crop(

        subclip, x_center=x_center, y_center=y_center, width=new_w, height=new_h

    )

    tacked_clips.append(cropped_clip)

reframed_clip = concatenate_videoclips(tacked_clips, method=”compose”)

reframed_clip.write_videofile(“output_video.mp4”)`

So basically I want to fix the face detection with ROI detection where it can detect the face and make that face and the body on to the frame and making sure that the speaker who is speaking is brought to the frame