#!/usr/bin/env python3 """ジェスチャー認識のデバッグスクリプト カメラから手を検出し、各指の状態と判定結果をリアルタイムで表示します。 """ import cv2 import numpy as np from rock_paper_scissors.detection import HandDetector, GestureDetector from rock_paper_scissors.detection.gesture_detector import ( LandmarkIndex, FINGER_EXTENDED_ANGLE_THRESHOLD, FINGER_CURLED_ANGLE_THRESHOLD, ) from rock_paper_scissors.game.states import Hand def calculate_finger_angle(landmarks: np.ndarray, mcp_idx: int, pip_idx: int, dip_idx: int) -> float: """指の曲がり角度を計算""" mcp = landmarks[mcp_idx][:2] pip = landmarks[pip_idx][:2] dip = landmarks[dip_idx][:2] v1 = mcp - pip v2 = dip - pip norm1 = np.linalg.norm(v1) norm2 = np.linalg.norm(v2) if norm1 < 1e-10 or norm2 < 1e-10: return 180.0 cos_angle = np.dot(v1, v2) / (norm1 * norm2) angle = np.arccos(np.clip(cos_angle, -1.0, 1.0)) return np.degrees(angle) def get_finger_debug_info(landmarks: np.ndarray) -> dict: """各指の詳細情報を取得""" info = {} # 各指の情報 fingers = { "index": (LandmarkIndex.INDEX_MCP, LandmarkIndex.INDEX_PIP, LandmarkIndex.INDEX_DIP, LandmarkIndex.INDEX_TIP), "middle": (LandmarkIndex.MIDDLE_MCP, LandmarkIndex.MIDDLE_PIP, LandmarkIndex.MIDDLE_DIP, LandmarkIndex.MIDDLE_TIP), "ring": (LandmarkIndex.RING_MCP, LandmarkIndex.RING_PIP, LandmarkIndex.RING_DIP, LandmarkIndex.RING_TIP), "pinky": (LandmarkIndex.PINKY_MCP, LandmarkIndex.PINKY_PIP, LandmarkIndex.PINKY_DIP, LandmarkIndex.PINKY_TIP), } for name, (mcp_idx, pip_idx, dip_idx, tip_idx) in fingers.items(): tip = landmarks[tip_idx] pip = landmarks[pip_idx] mcp = landmarks[mcp_idx] # 距離計算 tip_to_mcp = np.linalg.norm(tip[:2] - mcp[:2]) pip_to_mcp = np.linalg.norm(pip[:2] - mcp[:2]) ratio = tip_to_mcp / pip_to_mcp if pip_to_mcp > 0 else 0 # 角度計算 angle = calculate_finger_angle(landmarks, mcp_idx, pip_idx, dip_idx) # Y座標の比較 tip_above_pip = tip[1] < pip[1] # 判定 distance_extended = tip_above_pip and ratio > 0.9 angle_curled = angle < FINGER_CURLED_ANGLE_THRESHOLD # 最終判定（距離ベースで伸びていると判定されても、角度が小さければ閉じている） is_extended = distance_extended and not angle_curled info[name] = { "tip_y": tip[1], "pip_y": pip[1], "tip_above_pip": tip_above_pip, "distance_ratio": ratio, "angle": angle, "distance_extended": distance_extended, "angle_curled": angle_curled, "is_extended": is_extended, } # 親指 thumb_tip = landmarks[LandmarkIndex.THUMB_TIP] thumb_ip = landmarks[LandmarkIndex.THUMB_IP] thumb_mcp = landmarks[LandmarkIndex.THUMB_MCP] tip_to_mcp = np.linalg.norm(thumb_tip[:2] - thumb_mcp[:2]) ip_to_mcp = np.linalg.norm(thumb_ip[:2] - thumb_mcp[:2]) thumb_extended = tip_to_mcp > ip_to_mcp * 1.2 info["thumb"] = { "tip_to_mcp": tip_to_mcp, "ip_to_mcp": ip_to_mcp, "ratio": tip_to_mcp / ip_to_mcp if ip_to_mcp > 0 else 0, "is_extended": thumb_extended, } return info def main(): print("=" * 60) print("ジェスチャー認識デバッグツール") print("=" * 60) print(f"角度閾値: 伸びている >= {FINGER_EXTENDED_ANGLE_THRESHOLD}°, 曲がっている < {FINGER_CURLED_ANGLE_THRESHOLD}°") print("距離閾値: 0.9 (tip_to_mcp / pip_to_mcp)") print("'q' で終了") print("=" * 60) hand_detector = HandDetector() gesture_detector = GestureDetector() cap = cv2.VideoCapture(0) if not cap.isOpened(): print("カメラを開けませんでした") return try: while True: ret, frame = cap.read() if not ret: break # 左右反転（鏡像） frame = cv2.flip(frame, 1) # 手を検出 hand_data = hand_detector.detect(frame) if hand_data is not None: landmarks = hand_data.landmarks # ジェスチャー判定 gesture, confidence = gesture_detector.detect(landmarks) # デバッグ情報を取得 debug_info = get_finger_debug_info(landmarks) # 画面に表示 y_offset = 30 # ジェスチャー結果 gesture_text = f"Gesture: {gesture.value} (conf: {confidence:.2f})" color = (0, 255, 0) if gesture != Hand.UNKNOWN else (0, 0, 255) cv2.putText(frame, gesture_text, (10, y_offset), cv2.FONT_HERSHEY_SIMPLEX, 0.8, color, 2) y_offset += 35 # 拡張カウント extended_count = sum(1 for name in ["thumb", "index", "middle", "ring", "pinky"] if debug_info[name]["is_extended"]) cv2.putText(frame, f"Extended fingers: {extended_count}", (10, y_offset), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (255, 255, 255), 2) y_offset += 25 # 各指の情報 for finger_name in ["thumb", "index", "middle", "ring", "pinky"]: info = debug_info[finger_name] if finger_name == "thumb": text = f"{finger_name}: ext={info['is_extended']} (ratio={info['ratio']:.2f})" else: text = f"{finger_name}: ext={info['is_extended']} (ratio={info['distance_ratio']:.2f}, angle={info['angle']:.0f}°)" color = (0, 255, 0) if info["is_extended"] else (0, 0, 255) cv2.putText(frame, text, (10, y_offset), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 1) y_offset += 20 # ランドマークを描画 for i, lm in enumerate(landmarks): x = int(lm[0] * frame.shape[1]) y = int(lm[1] * frame.shape[0]) cv2.circle(frame, (x, y), 3, (0, 255, 255), -1) cv2.putText(frame, str(i), (x + 5, y), cv2.FONT_HERSHEY_SIMPLEX, 0.3, (255, 255, 255), 1) # コンソールにも出力（1秒に1回） import time if not hasattr(main, 'last_print') or time.time() - main.last_print > 1.0: main.last_print = time.time() print(f"\n--- {gesture.value} (conf: {confidence:.2f}) ---") for finger_name in ["thumb", "index", "middle", "ring", "pinky"]: info = debug_info[finger_name] ext_str = "○" if info["is_extended"] else "×" if finger_name == "thumb": print(f" {finger_name}: {ext_str} (ratio={info['ratio']:.2f})") else: print(f" {finger_name}: {ext_str} (ratio={info['distance_ratio']:.2f}, angle={info['angle']:.0f}°, tip_above={info['tip_above_pip']})") else: cv2.putText(frame, "No hand detected", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.8, (0, 0, 255), 2) cv2.imshow("Gesture Debug", frame) if cv2.waitKey(1) & 0xFF == ord('q'): break finally: cap.release() cv2.destroyAllWindows() hand_detector.close() if __name__ == "__main__": main()