Python | OpenCV 專案：車牌辨識應用

import cv2
print(cv2.__file__)           # 路徑裡不應該出現 "headless"
print(cv2.__version__)
print("imread" in dir(cv2))   # 應為 True

# recognize_plate.py
import cv2
import numpy as np
import easyocr

reader = easyocr.Reader(["en"], gpu=False)   # 台灣車牌為英數字，使用英文模型即可

def find_plate_candidates(img):
    """找出可能是車牌的矩形輪廓"""
    gray    = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    blur    = cv2.bilateralFilter(gray, 11, 17, 17)
    edges   = cv2.Canny(blur, 30, 200)

    contours, _ = cv2.findContours(edges, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
    contours     = sorted(contours, key=cv2.contourArea, reverse=True)[:20]

    candidates = []
    for c in contours:
        peri   = cv2.arcLength(c, True)
        approx = cv2.approxPolyDP(c, 0.018 * peri, True)
        if len(approx) == 4:
            candidates.append(approx)

    return candidates

def crop_plate(img, contour):
    """根據輪廓裁切車牌區域"""
    x, y, w, h = cv2.boundingRect(contour)
    return img[y:y+h, x:x+w], (x, y, w, h)

def preprocess_plate(plate_img):
    """車牌影像前處理"""
    gray    = cv2.cvtColor(plate_img, cv2.COLOR_BGR2GRAY)
    resized = cv2.resize(gray, (300, 80))
    _, binary = cv2.threshold(resized, 0, 255,
                              cv2.THRESH_BINARY + cv2.THRESH_OTSU)
    return binary

def recognize_plate(img_path):
    img = cv2.imread(img_path)
    if img is None:
        print("圖片讀取失敗")
        return

    candidates = find_plate_candidates(img)

    if not candidates:
        print("未找到車牌候選區域")
        return

    # 取面積最大的候選框
    plate_region, (x, y, w, h) = crop_plate(img, candidates[0])
    processed = preprocess_plate(plate_region)

    results = reader.readtext(processed, allowlist="ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789-")

    plate_text = " ".join([r[1] for r in results if r[2] > 0.3])
    print(f"辨識結果：{plate_text if plate_text else '無法辨識'}")

    # 繪製標記
    cv2.rectangle(img, (x, y), (x + w, y + h), (0, 255, 0), 3)
    cv2.putText(img, plate_text, (x, y - 10),
                cv2.FONT_HERSHEY_SIMPLEX, 1.0, (0, 255, 0), 2)

    cv2.imshow("Result", img)
    cv2.imshow("Plate", processed)
    cv2.waitKey(0)
    cv2.destroyAllWindows()

recognize_plate("assets/license_plate.jpg")

# perspective_transform.py
def four_point_transform(img, pts):
    """將四個角點矩形區域進行透視矯正"""
    pts = pts.reshape(4, 2).astype(np.float32)

    # 排序角點：左上、右上、右下、左下
    s    = pts.sum(axis=1)
    diff = np.diff(pts, axis=1)
    rect = np.array([
        pts[np.argmin(s)],    # 左上
        pts[np.argmin(diff)], # 右上
        pts[np.argmax(s)],    # 右下
        pts[np.argmax(diff)], # 左下
    ], dtype=np.float32)

    w = max(
        np.linalg.norm(rect[1] - rect[0]),
        np.linalg.norm(rect[2] - rect[3]),
    )
    h = max(
        np.linalg.norm(rect[3] - rect[0]),
        np.linalg.norm(rect[2] - rect[1]),
    )

    dst = np.array([[0, 0], [w, 0], [w, h], [0, h]], dtype=np.float32)
    M   = cv2.getPerspectiveTransform(rect, dst)
    return cv2.warpPerspective(img, M, (int(w), int(h)))

# scan_video.py
import cv2
import easyocr
import time

reader      = easyocr.Reader(["en"], gpu=False)
video_path  = "assets/traffic.mp4"   # 影片檔或 MJPEG 串流 URL 皆可
cap         = cv2.VideoCapture(video_path)
if not cap.isOpened():
    print(f"無法開啟影片：{video_path}")
    exit()

last_result = ""
last_bbox   = None
last_time   = 0
interval    = 1.0   # 每秒辨識一次，避免 OCR 頻繁呼叫拖慢幀率

print("影片辨識啟動，按 q 離開")

while True:
    ret, frame = cap.read()
    if not ret:
        print("影片播放完畢")
        break

    now = time.time()
    if now - last_time > interval:
        # EasyOCR 直接在整幀畫面偵測文字，比輪廓法更穩定
        results = reader.readtext(
            frame,
            allowlist="ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789-",
        )

        # 過濾：信心度 > 0.4、長寬比接近車牌比例（約 2:1 ~ 5:1）
        best = None
        for bbox, text, conf in results:
            if conf < 0.4:
                continue
            (tl, tr, br, bl) = bbox
            w = abs(tr[0] - tl[0])
            h = abs(bl[1] - tl[1])
            if h == 0 or not (1.5 < w / h < 6):
                continue
            if best is None or conf > best[2]:
                best = (bbox, text.replace(" ", ""), conf)

        if best:
            last_bbox, last_result, _ = best
        last_time = now

    if last_bbox is not None and last_result:
        tl = last_bbox[0]
        br = last_bbox[2]
        pt1 = (int(tl[0]), int(tl[1]))
        pt2 = (int(br[0]), int(br[1]))
        cv2.rectangle(frame, pt1, pt2, (0, 255, 0), 2)
        cv2.putText(frame, last_result, (pt1[0], pt1[1] - 10),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.8, (0, 255, 0), 2)

    cv2.imshow("License Plate Recognition", frame)
    if cv2.waitKey(30) & 0xFF == ord("q"):
        break

cap.release()
cv2.destroyAllWindows()