tabs/head_posture_detection.py

import tempfile
import cv2
import dlib
import numpy as np
from scipy.spatial import distance as dist
from imutils import face_utils
import gradio as gr

def detect_head_posture(video_path):
    detector = dlib.get_frontal_face_detector()
    predictor = dlib.shape_predictor("assets/models/shape_predictor_68_face_landmarks.dat")

    cap = cv2.VideoCapture(video_path)
    frame_width, frame_height = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)), int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
    
    with tempfile.NamedTemporaryFile(delete=False, suffix='.avi') as temp_file:
        out = cv2.VideoWriter(temp_file.name, cv2.VideoWriter_fourcc(*'XVID'), 20.0, (frame_width, frame_height))
        posture_data = []

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

            gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
            for rect in detector(gray, 0):
                shape = face_utils.shape_to_np(predictor(gray, rect))
                jaw_width = dist.euclidean(shape[1], shape[15])
                jaw_height = dist.euclidean(shape[8], (shape[1] + shape[15]) / 2)
                
                posture = "Upright" if jaw_height / jaw_width > 0.5 else "Slumped"
                posture_data.append(posture)

                for (x, y) in shape:
                    cv2.circle(frame, (x, y), 1, (0, 255, 0), -1)

            out.write(frame)

    cap.release()
    out.release()

    posture_type = max(set(posture_data), key=posture_data.count)
    return temp_file.name, posture_type

def create_head_posture_tab():
    with gr.Row():
        with gr.Column(scale=1):
            input_video = gr.Video(label="Input Video")
            with gr.Row():
                clear_btn = gr.Button("Clear")
                submit_btn = gr.Button("Analyze", elem_classes="submit")
        with gr.Column(scale=1, elem_classes="dl4"):
            output_video = gr.Video(label="Processed Video", elem_classes="video2")
            output_posture = gr.Label(label="Posture Type")
    
    submit_btn.click(detect_head_posture, inputs=input_video, outputs=[output_video, output_posture], queue=True)
    clear_btn.click(lambda: (None, None, None), outputs=[input_video, output_video, output_posture], queue=True)
    gr.Examples(["./assets/videos/fitness.mp4"], inputs=[input_video])