ytcollector 1.2.3__py3-none-any.whl → 1.2.4__py3-none-any.whl

ytcollector/__init__.py

@@ -22,7 +22,7 @@ from .analyzer import VideoAnalyzer, check_dependencies
 from .downloader import YouTubeDownloader
 from .cli import run, main as cli_main
 
-__version__ = "1.2.3"
+__version__ = "1.2.4"
 __all__ = [
     # 주요 클래스
     "VideoAnalyzer",

ytcollector/analyzer.py

@@ -32,7 +32,7 @@ try:
 except ImportError:
     USE_GPU = False
 
-from .config import LICENSE_PLATE_PATTERNS, LICENSE_PLATE_MODEL_PATH, YOLO_CONFIDENCE
+from .config import LICENSE_PLATE_PATTERNS, LICENSE_PLATE_MODEL_PATH, YOLO_CONFIDENCE, YOLO_HIGH_CONFIDENCE
 
 
 class VideoAnalyzer:
@@ -167,9 +167,9 @@ class VideoAnalyzer:
 
     def detect_license_plate(self, frame, texts=None):
         """
-        ROI 기반 번호판 감지 (최적화 버전)
+        ROI 기반 번호판 감지
         1. YOLO로 번호판 영역(ROI)을 먼저 찾음
-        2. 찾은 영역만 잘라서 OCR 수행 (속도 및 정확도 향상)
+        2. 고신뢰도(>=0.6)면 바로 인정, 저신뢰도면 OCR 패턴 매칭 필요
         """
         if not YOLO_AVAILABLE or frame is None:
             return False
@@ -177,59 +177,46 @@ class VideoAnalyzer:
         try:
             self._init_yolo()
             results = self.yolo_model(frame, verbose=False, conf=YOLO_CONFIDENCE)
-            
-            yolo_detected = False
-            roi_ocr_matched = False
 
             for r in results:
-                # 0: license plate
                 for box in r.boxes:
                     if box.cls == 0:
-                        yolo_detected = True
+                        conf = float(box.conf[0])
 
-                        # ROI 크기 필터링 (너무 작거나 큰 영역 제외)
+                        # ROI 크기 필터링
                         x1, y1, x2, y2 = map(int, box.xyxy[0])
                         roi_w, roi_h = x2 - x1, y2 - y1
                         frame_h, frame_w = frame.shape[:2]
                         roi_area_ratio = (roi_w * roi_h) / (frame_w * frame_h)
 
-                        # ROI가 전체 화면의 0.1% ~ 20% 사이여야 함
                         if roi_area_ratio < 0.001 or roi_area_ratio > 0.2:
                             continue
 
-                        # ROI 크로핑 및 타겟 OCR
+                        # 고신뢰도면 바로 인정
+                        if conf >= YOLO_HIGH_CONFIDENCE:
+                            return True
+
+                        # 저신뢰도면 OCR 패턴 매칭 필요
                         h, w = frame.shape[:2]
-                        # 패딩 10% 추가
-                        pad_w = int((x2 - x1) * 0.1)
-                        pad_h = int((y2 - y1) * 0.1)
-                        
+                        pad_w = int((x2 - x1) * 0.15)
+                        pad_h = int((y2 - y1) * 0.15)
+
                         crop_x1 = max(0, x1 - pad_w)
                         crop_y1 = max(0, y1 - pad_h)
                         crop_x2 = min(w, x2 + pad_w)
                         crop_y2 = min(h, y2 + pad_h)
-                        
+
                         roi = frame[crop_y1:crop_y2, crop_x1:crop_x2]
                         if roi.size > 0:
-                            # ROI에 대해서만 OCR 실행
                             roi_texts = self.detect_text(roi)
                             if roi_texts:
-                                combined_roi = "".join([re.sub(r'[^0-9가-힣]', '', t) for t in roi_texts])
+                                combined = "".join([re.sub(r'[^0-9가-힣A-Za-z]', '', t) for t in roi_texts])
                                 for pattern in LICENSE_PLATE_PATTERNS:
-                                    # 개별 텍스트 및 결합 텍스트 확인
                                     if any(re.search(pattern, re.sub(r'[^0-9가-힣]', '', t)) for t in roi_texts) or \
-                                       re.search(pattern, combined_roi):
-                                        roi_ocr_matched = True
-                                        break
-                        
-                        if roi_ocr_matched: break
-                if roi_ocr_matched: break
-
-            # 최종 판정: YOLO 감지 + OCR 패턴 매칭 둘 다 만족해야 함 (오탐지 방지)
-            if yolo_detected and roi_ocr_matched:
-                return True
-                
+                                       re.search(pattern, combined):
+                                        return True
+
         except Exception as e:
-            # print(f"  ⚠ 번호판 ROI 분석 에러: {e}")
             pass
 
         return False

ytcollector/config.py

@@ -81,7 +81,8 @@ BLACKLIST_KEYWORDS = {
 }
 
 # YOLO 설정
-YOLO_CONFIDENCE = 0.5  # 오탐지 방지를 위해 신뢰도 상향
+YOLO_CONFIDENCE = 0.3  # 감지율 향상을 위해 낮춤
+YOLO_HIGH_CONFIDENCE = 0.6  # 고신뢰도 (OCR 없이 바로 인정)
 
 # 번호판 정규식 패턴 (한국 자동차 번호판 중심)
 LICENSE_PLATE_PATTERNS = [

{ytcollector-1.2.3.dist-info → ytcollector-1.2.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ytcollector
-Version: 1.2.3
+Version: 1.2.4
 Summary: YouTube 콘텐츠 수집기 - 얼굴, 번호판, 타투, 텍스트 감지
 Author: YTCollector Team
 License: MIT

ytcollector-1.2.4.dist-info/RECORD

@@ -0,0 +1,13 @@
+ytcollector/__init__.py,sha256=Z6wl96XcqT160HYKVglJiEAxAnaVjUYC--KrBbAMmmY,1152
+ytcollector/analyzer.py,sha256=tUD3-SrcyXuSAkfOUERM1bd-cSjZiMfa51Yd66njkLY,13693
+ytcollector/cli.py,sha256=aHF4EuQRPLKh65lnkI_dZ0ResztlVjpHlS5iHfzmpig,5577
+ytcollector/config.py,sha256=2v4tFn1LX7aljsoFMdO8woq_ptNTsvTEy5_5WssviC0,3706
+ytcollector/dataset_builder.py,sha256=nfArEwszoCln48n3T0Eff_4OOaYv8FF0YH8cARBGMWQ,2608
+ytcollector/downloader.py,sha256=TeC6agUmSPHZSZ9jdoc42i8i_NobzTEkoRtAIgW80kI,14544
+ytcollector/utils.py,sha256=6XDif-e3GbMHmUvTsBT0YblxNxYnS-2I8HnmjMBZs-M,4254
+ytcollector/models/license_plate_detector.pt,sha256=jsOyVKbIdhDwN6kJV0Ysr6EanAMiTjOijGodGsKsUbA,6241454
+ytcollector-1.2.4.dist-info/METADATA,sha256=9afCmbZ1kxNwM1hbXuvulER6eWKIy_xAtvUVApEDjzg,7319
+ytcollector-1.2.4.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+ytcollector-1.2.4.dist-info/entry_points.txt,sha256=waiVuSJJYt-6_DAal-T4JkHgejo7wKYLdKrEI7tZ-ms,127
+ytcollector-1.2.4.dist-info/top_level.txt,sha256=wozNyCUm0eMOm-9U81yTql6oGaM2O5rWVBXDb93zzyQ,12
+ytcollector-1.2.4.dist-info/RECORD,,

ytcollector-1.2.3.dist-info/RECORD

@@ -1,13 +0,0 @@
-ytcollector/__init__.py,sha256=P4AIOCG2glB95wwN9QhUuF0HsgVxy6JyjTvkZKuQQXg,1152
-ytcollector/analyzer.py,sha256=zgtcw6pSCzJwa4NJ04XFJlLaoRt8CwXGxwuL9o8eduM,14414
-ytcollector/cli.py,sha256=aHF4EuQRPLKh65lnkI_dZ0ResztlVjpHlS5iHfzmpig,5577
-ytcollector/config.py,sha256=lpzYaODbOErUni_oAJ_zU0dpIQzy8gJDz-Cisrqf1tE,3646
-ytcollector/dataset_builder.py,sha256=nfArEwszoCln48n3T0Eff_4OOaYv8FF0YH8cARBGMWQ,2608
-ytcollector/downloader.py,sha256=TeC6agUmSPHZSZ9jdoc42i8i_NobzTEkoRtAIgW80kI,14544
-ytcollector/utils.py,sha256=6XDif-e3GbMHmUvTsBT0YblxNxYnS-2I8HnmjMBZs-M,4254
-ytcollector/models/license_plate_detector.pt,sha256=jsOyVKbIdhDwN6kJV0Ysr6EanAMiTjOijGodGsKsUbA,6241454
-ytcollector-1.2.3.dist-info/METADATA,sha256=u3Xz0naQgl9F-ZFOM7vw4fu25KxR28KfOPuz6eTzGkI,7319
-ytcollector-1.2.3.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
-ytcollector-1.2.3.dist-info/entry_points.txt,sha256=waiVuSJJYt-6_DAal-T4JkHgejo7wKYLdKrEI7tZ-ms,127
-ytcollector-1.2.3.dist-info/top_level.txt,sha256=wozNyCUm0eMOm-9U81yTql6oGaM2O5rWVBXDb93zzyQ,12
-ytcollector-1.2.3.dist-info/RECORD,,