Editing Openai/692fcfd8-4b44-800f-a184-fa098175a6d5 (section)

==== : ====

請整份複製貼上：

<syntaxhighlight lang="python"># hand_gesture_1_10.py
=== Python 3.7 版：手勢 1～10 辨識＋教學面板＋國語／台語語音 ===

from cvzone.HandTrackingModule import HandDetector
import cv2
import math  # 雖然不用 math.dist，但保留 math 以備後續若要加東西
import os
import time
import pygame
import numpy as np

=== ========= 小工具：2D 距離（取代 math.dist） ========= ===
def euclidean(a, b):
    """
    計算 2D 歐幾里得距離
    a, b: [x, y, (z)] 形式的 list 或 tuple，這裡只用前兩個值
    """
    return ((a[0] - b[0]) ''' 2 + (a[1] - b[1]) ''' 2) ** 0.5

=== ========== 初始設定 ========== ===
=== 攝影機：0 通常是內建鏡頭；如有外接可換成 1 ===
cap = cv2.VideoCapture(0)  # 若遇到裝置問題，可試 cv2.VideoCapture(0, cv2.CAP_DSHOW)

=== 手部偵測器（最多 2 隻手，為了辨識 10） ===
detector = HandDetector(detectionCon=0.7, maxHands=2)

=== 初始化 pygame 音效系統 ===
pygame.mixer.init()

=== 台語 POJ 對照 ===
tai_map = {
    1: "it",
    2: "nn̄g",
    3: "saⁿ",
    4: "sì",
    5: "gō͘",
    6: "la̍k",
    7: "chhit",
    8: "peh",
    9: "káu",
    10: "chap",
}

=== 中文數字對照 ===
zh_map = {
    1: "一",
    2: "二",
    3: "三",
    4: "四",
    5: "五",
    6: "六",
    7: "七",
    8: "八",
    9: "九",
    10: "十",
}

=== ========= 讀取手勢示意圖 ========= ===
gesture_images = {}
for n in range(1, 11):
    path = os.path.join("gestures", f"{n}.png")
    if os.path.exists(path):
        img_g = cv2.imread(path)
        if img_g is not None:
            gesture_images[n] = img_g
    else:
        print("[提示] 找不到手勢示意圖：{}".format(path))

=== ========= 播放語音 ========= ===
def play_audio(number):
    """
    播放對應數字的國語 + 台語語音
    需要 audio/zh_N.mp3, audio/tai_N.mp3 已存在
    """
    for lang in ["zh", "tai"]:
        fn = os.path.join("audio", "{}_{}.mp3".format(lang, number))
        if os.path.exists(fn):
            try:
                pygame.mixer.music.load(fn)
                pygame.mixer.music.play()
                # 簡單阻塞等待播放完畢
                while pygame.mixer.music.get_busy():
                    time.sleep(0.05)
            except Exception as e:
                print("[錯誤] 播放 {} 失敗：{}".format(fn, e))
        else:
            print("[提示] 找不到音檔：{}".format(fn))

=== ========= 手勢判斷（單手 1～9） ========= ===
def classify_single_hand_number(hand, fingers):
    """
    根據單手的 fingersUp 結果判斷數字 1~9

    hand: 單手資訊 (包含 lmList, bbox, type 等)
    fingers: [thumb, index, middle, ring, pinky] (0/1)
    回傳 (number, label) 或 (None, "None")
    """
    pattern = tuple(fingers)

    # 固定手勢模式（不含「只伸食指」，那留給 1/9 特判）
    mapping = {
        (0, 1, 1, 0, 0): (2, "2"),      # 食 + 中
        (1, 1, 1, 0, 0): (3, "3"),      # 拇 + 食 + 中
        (0, 1, 1, 1, 1): (4, "4"),      # 四指伸、拇指收
        (1, 1, 1, 1, 1): (5, "5"),      # 全部伸出
        (1, 0, 0, 0, 1): (6, "6"),      # 拇 + 小（shaka）
        (1, 1, 0, 0, 1): (7, "7"),      # 拇 + 食 + 小
        (1, 1, 0, 0, 0): (8, "8"),      # 拇 + 食（手槍）
    }

    if pattern in mapping:
        return mapping[pattern]

    # --- 特判：1 或 9（只伸食指）---
    if pattern == (0, 1, 0, 0, 0):
        lm_list = hand["lmList"]
        # 食指：MCP = 5, PIP = 6, DIP = 7, TIP = 8
        mcp = lm_list[5]
        pip = lm_list[6]
        dip = lm_list[7]
        tip = lm_list[8]

        # 直線距離 MCP→TIP
        dist_mcp_tip = euclidean(mcp, tip)

        # 折線距離 MCP→PIP→DIP→TIP
        dist_mcp_pip = euclidean(mcp, pip)
        dist_pip_dip = euclidean(pip, dip)
        dist_dip_tip = euclidean(dip, tip)
        poly_len = dist_mcp_pip + dist_pip_dip + dist_dip_tip

        # 粗略規則：如果「折線 / 直線」比例大於 1.25，視為彎曲很多 = 9
        if dist_mcp_tip > 0 and poly_len / dist_mcp_tip > 1.25:
            return (9, "9")
        else:
            return (1, "1")

    return (None, "None")

=== ========= 判斷整體數字（包含 10） ========= ===
def classify_number(hands):
    """
    根據全部手部資料判斷數字 1~10
    回傳 (number, label, bbox_for_draw)
    """
    if not hands:
        return None, "No Hand", None

    # 先判斷「10」：兩手都比 1
    if len(hands) >= 2:
        nums = []
        for h in hands[:2]:
            fingers = detector.fingersUp(h)
            n, _ = classify_single_hand_number(h, fingers)
            nums.append(n)

        if nums[0] == 1 and nums[1] == 1:
            bbox10 = hands[0]["bbox"]
            return 10, "10", bbox10

    # 否則只看第一隻手的 1~9
    main_hand = hands[0]
    bbox = main_hand["bbox"]
    fingers = detector.fingersUp(main_hand)
    number, label = classify_single_hand_number(main_hand, fingers)

    if number is None:
        return None, "Unknown", bbox
    else:
        return number, label, bbox

=== ========= 主迴圈 ========= ===
last_number = None  # 上一次辨識結果，避免每幀都重播聲音

while cap.isOpened():
    success, img = cap.read()
    if not success:
        break

    # 偵測手部
    hands, img = detector.findHands(img)

    # 判斷數字
    number, label, bbox = classify_number(hands)

    # 原始畫面大小
    h, w, _ = img.shape

    # 建立右側教學面板（白底）
    panel_width = 320
    panel = 255 * np.ones((h, panel_width, 3), dtype="uint8")

    # ===== 在教學面板顯示內容 =====
    if number is not None and number in tai_map:
        # 顯示手勢示意圖（若存在）
        if number in gesture_images:
            g_img = gesture_images[number]
            gh, gw, _ = g_img.shape

            # 讓示意圖高度約畫面 60%，寬度約 panel 90%
            scale = min((h '' 0.6) / float(gh), (panel_width '' 0.9) / float(gw))
            new_w = int(gw * scale)
            new_h = int(gh * scale)
            g_resized = cv2.resize(g_img, (new_w, new_h))

            # 貼到 panel 置中（上方）
            x_offset = (panel_width - new_w) // 2
            y_offset = 20
            panel[y_offset:y_offset + new_h, x_offset:x_offset + new_w] = g_resized

        # 文字資訊（下半部）
        text_y = int(h * 0.75)
        cv2.putText(panel, str(number),
                    (30, text_y),
                    cv2.FONT_HERSHEY_SIMPLEX, 1.2, (0, 0, 0), 2)
        cv2.putText(panel, "中文：{}".format(zh_map[number]),
                    (30, text_y + 30),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 0), 2)
        cv2.putText(panel, "台語：{}".format(tai_map[number]),
                    (30, text_y + 60),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 0), 2)
    else:
        cv2.putText(panel, "請比出 1~10 的手勢",
                    (20, h // 2),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.7, (128, 128, 128), 2)

    # 在原影像畫偵測框與標籤
    if bbox is not None and number is not None:
        x, y, bw, bh = bbox
        cv2.rectangle(img, (x, y), (x + bw, y + bh), (0, 255, 0), 2)
        cv2.putText(img, label,
                    (x, y - 10),
                    cv2.FONT_HERSHEY_PLAIN, 2, (0, 255, 0), 2)

    # 將攝影機畫面與教學面板合併（左右排）
    panel = cv2.resize(panel, (panel_width, h))
    combined = cv2.hconcat([img, panel])

    cv2.imshow("Hand Number Teaching (1-10) - Python 3.7", combined)

    # 若辨識數字有變化 → 播放語音
    if number is not None and number in tai_map and number != last_number:
        last_number = number
        print("辨識到數字：", number)
        play_audio(number)

    # 按 q 離開
    if cv2.waitKey(1) & 0xFF == ord("q"):
        break

cap.release()
cv2.destroyAllWindows()

</syntaxhighlight>