'commit'

Apps/TeLaiDian/ReadImageKit.py

@@ -10,7 +10,7 @@ if project_root not in sys.path:
     sys.path.append(project_root)
 
 from Util.VLMKit import VLMKit
-from Apps.TeLaiDian.Kit import draw_rectangles, detect_cards_cv, setup_logger
+from Apps.TeLaiDian.Kit import draw_rectangles, detect_cards_cv, setup_logger, read_image
 from Apps.TeLaiDian.Config.Setting import SAFE_EXCLUDE_RATIO, BOTTOM_SAFE_EXCLUDE_RATIO
 
 # 初始化日志
@@ -224,16 +224,25 @@ class ReadImageKit:
                 for res in vlm_list:
                     p = res.get("point")
                     if p and len(p) == 2:
-                        # 如果坐标小于 1000，认为是归一化坐标，需要转换
+                        # 1. 坐标转换逻辑
+                        actual_p = p
                         if p[0] <= 1000 and p[1] <= 1000:
                             if w == 0:
-                                import cv2
+                                img = read_image(image_path)
-                                img = cv2.imread(image_path)
                                 if img is not None:
                                     h, w = img.shape[:2]
                                 else:
                                     w, h = 1080, 2400 # 兜底
-                            res["point"] = [int(p[0] * w / 1000), int(p[1] * h / 1000)]
+                            actual_p = [int(p[0] * w / 1000), int(p[1] * h / 1000)]
+                        
+                        # 2. 坐标安全过滤：忽略页面上半部分的误点击（通常是 Logo 或广告）
+                        # 即使 VLM 没听话，我们也在这里硬性过滤
+                        y_threshold = h * 0.4 if h > 0 else 800
+                        if actual_p[1] < y_threshold:
+                            logger.warning(f"过滤掉可能的误点击点 (y={actual_p[1]} < {y_threshold}): {res.get('name')}")
+                            continue
+                            
+                        res["point"] = actual_p
                     final_stations.append(res)
                 
             return final_stations