app/pose_processor.py

import gc

import cv2
from app.llm import ensure_model_pose_exists
import mediapipe as mp
from mediapipe.tasks.python import vision
from mediapipe.tasks.python import BaseOptions
from app.config import FRAME_SKIP_POSE, POSE_MODEL_PATH
from app.logger import log

def extract_body_signals(video_path, logger=None):
    log(f"Extraindo sinais corporais de {video_path}...", logger=logger)
    ensure_model_pose_exists(logger=logger)

    cap = cv2.VideoCapture(video_path)
    log("Vídeo carregado com sucesso.", logger=logger)

    movements = []
    prev_wrist = None
    frame_count = 0

    # Essa parte é necessária para contornar um problema do MediaPipe que não aceita modelos carregados diretamente do disco.
    # Erro:Unable to open file at C:\Pos_IA\hf_backend\.venv-311\Lib\site-packages/C:\Pos_IA\hf_backend\models\pose_landmarker.task, errno=22
    #options = vision.PoseLandmarkerOptions(
    #    base_options=BaseOptions(model_asset_path=str(POSE_MODEL_PATH)),
    #    num_poses=1
    #)

    log(f"Configurando detector de pose e carregando modelo em {POSE_MODEL_PATH}", logger=logger)
    with open(POSE_MODEL_PATH, "rb") as f:
        model_bytes = f.read()

    # 🔧 Configurar detector
    options = vision.PoseLandmarkerOptions(
        base_options=BaseOptions(model_asset_buffer=model_bytes),
        num_poses=1
    )

    detector = vision.PoseLandmarker.create_from_options(options)

    log("Detector de pose criado com sucesso.", logger=logger)

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

        if frame_count % FRAME_SKIP_POSE != 0:
            frame_count += 1
            continue

        rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
        
        # converter para MediaPipe Image
        mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=rgb)

        result = detector.detect(mp_image)

        if result.pose_landmarks:
            lm = result.pose_landmarks[0]

            # RIGHT_WRIST = índice 16
            wrist = (lm[16].x, lm[16].y)

            if prev_wrist:
                dist = ((wrist[0] - prev_wrist[0])**2 + (wrist[1] - prev_wrist[1])**2)**0.5
                movements.append(dist)

            prev_wrist = wrist

        frame_count += 1

    cap.release()
    detector.close()
    del detector
    gc.collect()
    try:
        cv2.destroyAllWindows()
    except:
        pass

    avg_movement = sum(movements)/len(movements) if movements else 0

    log(f"Movimento corporal: {avg_movement}", logger=logger)
    log("Sinais corporais extraídos com sucesso.", logger=logger)

    return {
        "body_movement": avg_movement
    }