You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
149 lines
5.7 KiB
149 lines
5.7 KiB
import cv2
|
|
import numpy as np
|
|
import mediapipe as mp
|
|
from mediapipe.tasks import python
|
|
from mediapipe.tasks.python import vision
|
|
import time
|
|
import random
|
|
import winsound # Windows用の音源（Mac/Linuxの場合は別ライブラリが必要）
|
|
import csv
|
|
from datetime import datetime
|
|
|
|
# --- 設定 ---
|
|
MODEL_PATH = 'face_landmarker.task'
|
|
MIN_SOUND_INTERVAL = 5.0 # 音を鳴らす最小間隔（秒）
|
|
MAX_SOUND_INTERVAL = 10.0 # 音を鳴らす最大間隔（秒）
|
|
HISTORY_LENGTH = 100 # 画面に表示するグラフのデータ点数
|
|
|
|
# --- グローバル変数 ---
|
|
last_sound_time = time.time()
|
|
next_interval = random.uniform(MIN_SOUND_INTERVAL, MAX_SOUND_INTERVAL)
|
|
sound_triggered_flag = False
|
|
|
|
# データ保存用バッファ [(記録日時, 経過時間[s], Yaw値, 音フラグ)]
|
|
data_log = []
|
|
yaw_history = []
|
|
start_time = time.time()
|
|
|
|
# --- MediaPipe セットアップ ---
|
|
BaseOptions = python.BaseOptions
|
|
FaceLandmarker = vision.FaceLandmarker
|
|
FaceLandmarkerOptions = vision.FaceLandmarkerOptions
|
|
VisionRunningMode = vision.RunningMode
|
|
|
|
options = FaceLandmarkerOptions(
|
|
base_options=BaseOptions(model_asset_path=MODEL_PATH),
|
|
running_mode=VisionRunningMode.VIDEO,
|
|
output_facial_transformation_matrixes=True,
|
|
num_faces=1
|
|
)
|
|
|
|
cap = cv2.VideoCapture(0)
|
|
|
|
print("プログラムを開始します。'q' キーで終了し、CSVへ保存します。")
|
|
|
|
with FaceLandmarker.create_from_options(options) as landmarker:
|
|
while cap.isOpened():
|
|
success, frame = cap.read()
|
|
if not success:
|
|
break
|
|
|
|
current_time = time.time()
|
|
elapsed_time = current_time - start_time
|
|
|
|
# 1. ランダムなタイミングで音を鳴らす制御
|
|
is_sound_now = 0
|
|
if current_time - last_sound_time > next_interval:
|
|
# 2000Hzの音を200ミリ秒鳴らす（スレッドをブロックしないよう、短い音に設定）
|
|
winsound.Beep(2000, 200)
|
|
|
|
print(f"[{datetime.now().strftime('%H:%M:%S')}] ♪ 音を鳴らしました")
|
|
last_sound_time = current_time
|
|
next_interval = random.uniform(MIN_SOUND_INTERVAL, MAX_SOUND_INTERVAL)
|
|
is_sound_now = 1 # CSV記録用のフラグ
|
|
|
|
# 2. 顔検出とYaw値の計算
|
|
rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
|
frame_timestamp_ms = int(cap.get(cv2.CAP_PROP_POS_MSEC))
|
|
if frame_timestamp_ms == 0:
|
|
frame_timestamp_ms = int(elapsed_time * 1000)
|
|
|
|
mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=rgb_frame)
|
|
result = landmarker.detect_for_video(mp_image, frame_timestamp_ms)
|
|
|
|
yaw = None
|
|
if result.facial_transformation_matrixes:
|
|
for matrix in result.facial_transformation_matrixes:
|
|
rot_mat = matrix[:3, :3]
|
|
angles, _, _, _, _, _ = cv2.RQDecomp3x3(rot_mat)
|
|
yaw = angles[1] # 左右の向き (Yaw)
|
|
|
|
# 3. データの記録 (顔が検出できない場合は None または 0 として記録)
|
|
current_yaw = yaw if yaw is not None else 0.0
|
|
|
|
# メモリに時系列データを蓄積
|
|
data_log.append([
|
|
datetime.now().strftime('%Y-%m-%d %H:%M:%S.%f')[:-3],
|
|
round(elapsed_time, 3),
|
|
round(current_yaw, 2),
|
|
is_sound_now
|
|
])
|
|
|
|
# リアルタイムグラフ表示用の履歴更新
|
|
yaw_history.append(current_yaw)
|
|
if len(yaw_history) > HISTORY_LENGTH:
|
|
yaw_history.pop(0)
|
|
|
|
# 4. 視覚化（カメラ映像へのテキスト描画）
|
|
if yaw is not None:
|
|
cv2.putText(frame, f"Yaw: {current_yaw:.1f}", (20, 40),
|
|
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2)
|
|
else:
|
|
cv2.putText(frame, "Face Lost", (20, 40),
|
|
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
|
|
|
|
if is_sound_now:
|
|
cv2.putText(frame, "SOUND DETECTED!", (20, 80),
|
|
cv2.FONT_HERSHEY_SIMPLEX, 1.0, (0, 0, 255), 3)
|
|
|
|
# 5. 簡易リアルタイムグラフ（波形）の描画
|
|
# 画面下部に黒いグラフエリアを作成
|
|
graph_h, graph_w = 150, frame.shape[1]
|
|
graph = np.zeros((graph_h, graph_w, 3), dtype=np.uint8)
|
|
cv2.line(graph, (0, graph_h // 2), (graph_w, graph_h // 2), (100, 100, 100), 1) # 中心線(0度)
|
|
|
|
if len(yaw_history) > 1:
|
|
point_interval = graph_w / HISTORY_LENGTH
|
|
for i in range(1, len(yaw_history)):
|
|
# 角度をグラフのY座標にマッピング（上下に最大±45度を想定）
|
|
y1 = int(graph_h / 2 - (yaw_history[i - 1] * (graph_h / 90)))
|
|
y2 = int(graph_h / 2 - (yaw_history[i] * (graph_h / 90)))
|
|
|
|
x1 = int((i - 1) * point_interval)
|
|
x2 = int(i * point_interval)
|
|
|
|
# 範囲外のクリッピング
|
|
y1 = max(0, min(graph_h - 1, y1))
|
|
y2 = max(0, min(graph_h - 1, y2))
|
|
|
|
cv2.line(graph, (x1, y1), (x2, y2), (0, 255, 255), 2)
|
|
|
|
# カメラ映像とグラフを縦に連結して表示
|
|
combined_view = np.vstack((frame, graph))
|
|
cv2.imshow('Face Orientation & Timeline', combined_view)
|
|
|
|
if cv2.waitKey(1) & 0xFF == ord('q'):
|
|
break
|
|
|
|
cap.release()
|
|
cv2.destroyAllWindows()
|
|
|
|
# --- 6. CSVファイルへのデータ書き出し ---
|
|
filename = f"turning_detection_{datetime.now().strftime('%Y%m%d_%H%M%S')}.csv"
|
|
with open(filename, 'w', newline='', encoding='utf-8') as f:
|
|
writer = csv.writer(f)
|
|
# ヘッダー行
|
|
writer.writerow(["Timestamp", "Elapsed_Seconds", "Yaw_Angle", "Sound_Triggered"])
|
|
writer.writerows(data_log)
|
|
|
|
print(f"\nデータが正常に保存されました: {filename}") |