#!/usr/bin/env python3 # -*- coding: utf-8 -*- """ 单张图片快速测试：三角形四角标记识别 + 单应性落点 + PnP 估距 用法（在板子上）： python3 test/test_triangle_one_image.py --image /root/phot/xxx.jpg --out /root/phot/tri_out.jpg 调参对比（不改代码，临时覆盖 config.TRIANGLE_*）： python3 test/test_triangle_one_image.py --image /root/phot/xxx.jpg --preset shadow python3 test/test_triangle_one_image.py --image /root/phot/xxx.jpg --max-interior-gray 160 --min-dark-ratio 0.20 """ import argparse import json import os import time from typing import Any, Dict, Tuple import cv2 import numpy as np import config import triangle_target as tri_mod from triangle_target import ( detect_triangle_markers, load_camera_from_xml, load_triangle_positions, try_triangle_scoring, ) def _apply_overrides(args) -> None: # 预设：阴影/低对比度场景更宽松（尽量保持速度：不启 CLAHE） if args.preset == "shadow": setattr(config, "TRIANGLE_ENABLE_CLAHE_FALLBACK", False) setattr(config, "TRIANGLE_MIN_CONTRAST_DIFF", 0) setattr(config, "TRIANGLE_MAX_INTERIOR_GRAY", 160) setattr(config, "TRIANGLE_DARK_PIXEL_GRAY", 160) setattr(config, "TRIANGLE_MIN_DARK_RATIO", 0.20) # adaptive 只在 Otsu 失败时尝试，保持尝试次数很少 setattr(config, "TRIANGLE_ADAPTIVE_BLOCK_SIZES", (21,)) # 手动覆盖（优先级高于 preset） if args.max_interior_gray is not None: setattr(config, "TRIANGLE_MAX_INTERIOR_GRAY", int(args.max_interior_gray)) if args.dark_pixel_gray is not None: setattr(config, "TRIANGLE_DARK_PIXEL_GRAY", int(args.dark_pixel_gray)) if args.min_dark_ratio is not None: setattr(config, "TRIANGLE_MIN_DARK_RATIO", float(args.min_dark_ratio)) if args.min_contrast_diff is not None: setattr(config, "TRIANGLE_MIN_CONTRAST_DIFF", int(args.min_contrast_diff)) if args.detect_scale is not None: setattr(config, "TRIANGLE_DETECT_SCALE", float(args.detect_scale)) if args.adaptive_blocks is not None: bs = tuple(int(x) for x in args.adaptive_blocks.split(",") if x.strip()) setattr(config, "TRIANGLE_ADAPTIVE_BLOCK_SIZES", bs) def _dump_config() -> Dict[str, Any]: keys = [ "TRIANGLE_DETECT_SCALE", "TRIANGLE_SIZE_RANGE", "TRIANGLE_MAX_INTERIOR_GRAY", "TRIANGLE_DARK_PIXEL_GRAY", "TRIANGLE_MIN_DARK_RATIO", "TRIANGLE_MIN_CONTRAST_DIFF", "TRIANGLE_ADAPTIVE_BLOCK_SIZES", "TRIANGLE_MAX_FILTERED_FOR_COMBO", "TRIANGLE_EARLY_EXIT_CANDIDATES", "TRIANGLE_ENABLE_CLAHE_FALLBACK", ] out = {} for k in keys: out[k] = getattr(config, k, None) return out def _draw_tri_debug(img_bgr: np.ndarray, tri: Dict[str, Any]) -> np.ndarray: out = img_bgr.copy() markers = tri.get("markers") or [] # 画三角形轮廓 + center + id for m in markers: corners = np.array(m.get("corners", []), dtype=np.int32) if corners.size == 0: continue cv2.polylines(out, [corners], True, (0, 255, 0), 2) c = m.get("center") or (corners[:, 0].mean(), corners[:, 1].mean()) cx, cy = int(c[0]), int(c[1]) cv2.circle(out, (cx, cy), 4, (0, 0, 255), -1) mid = m.get("id", "?") cv2.putText(out, f"T{mid}", (cx - 18, cy - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.55, (0, 255, 0), 1) # 若有 homography，画靶心（把 (0,0) 反投影到图像） H = tri.get("homography") if H is not None: try: H = np.array(H, dtype=np.float64) H_inv = np.linalg.inv(H) c_img = cv2.perspectiveTransform(np.array([[[0.0, 0.0]]], dtype=np.float32), H_inv)[0][0] ocx, ocy = int(c_img[0]), int(c_img[1]) cv2.circle(out, (ocx, ocy), 5, (0, 0, 255), -1) cv2.circle(out, (ocx, ocy), 10, (0, 0, 255), 1) except Exception: pass # 叠加结果信息 lines = [] if tri.get("ok"): lines.append("tri_ok=True") if tri.get("dx_cm") is not None and tri.get("dy_cm") is not None: lines.append(f"dx,dy=({tri['dx_cm']:.2f},{tri['dy_cm']:.2f})cm") if tri.get("distance_m") is not None: lines.append(f"dist={float(tri['distance_m']):.2f}m") else: lines.append("tri_ok=False") y0 = 22 for i, t in enumerate(lines): cv2.putText(out, t, (10, y0 + i * 18), cv2.FONT_HERSHEY_SIMPLEX, 0.55, (0, 255, 0), 1) return out def main(): ap = argparse.ArgumentParser() ap.add_argument("--image", required=True, help="输入图片路径（jpg/png）") ap.add_argument("--out", default="", help="输出标注图片路径（可选）") ap.add_argument("--laser-x", type=int, default=-1, help="激光点 x（像素），默认用图像中心") ap.add_argument("--laser-y", type=int, default=-1, help="激光点 y（像素），默认用图像中心") ap.add_argument("--preset", choices=["", "shadow"], default="", help="调参预设（shadow=阴影更鲁棒，不启 CLAHE）") ap.add_argument("--max-interior-gray", type=int, default=None) ap.add_argument("--dark-pixel-gray", type=int, default=None) ap.add_argument("--min-dark-ratio", type=float, default=None) ap.add_argument("--min-contrast-diff", type=int, default=None) ap.add_argument("--detect-scale", type=float, default=None) ap.add_argument("--adaptive-blocks", default=None, help="例如: 11,21 ；为空表示不改") ap.add_argument("--verbose", action="store_true", help="输出更多检测阶段信息") args = ap.parse_args() _apply_overrides(args) # triangle_target.py 的日志默认写到 logger_manager；在离线脚本里 logger 可能未初始化。 # verbose 模式下把 _log 重定向为 print，方便直接看到诊断信息。 if args.verbose: try: tri_mod._log = lambda msg: print(str(msg)) except Exception: pass img_bgr = cv2.imread(args.image, cv2.IMREAD_COLOR) if img_bgr is None: raise SystemExit(f"读图失败：{args.image}") # triangle_target.try_triangle_scoring 约定输入为 RGB；OpenCV imread 返回 BGR img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB) h, w = img_bgr.shape[:2] if args.laser_x >= 0 and args.laser_y >= 0: laser_point = (int(args.laser_x), int(args.laser_y)) else: laser_point = (w // 2, h // 2) K, dist = load_camera_from_xml(getattr(config, "CAMERA_CALIB_XML", "")) pos = load_triangle_positions(getattr(config, "TRIANGLE_POSITIONS_JSON", "")) print("[tri-test] image:", args.image, "shape:", (h, w)) print("[tri-test] laser_point:", laser_point) print("[tri-test] calib_ok:", bool(K is not None and dist is not None), "pos_ok:", bool(pos)) print("[tri-test] config:", json.dumps(_dump_config(), ensure_ascii=False)) # 先单独跑一次三角形候选检测，便于区分“没找到候选” vs “找到候选但评分/单应性失败” scale = float(getattr(config, "TRIANGLE_DETECT_SCALE", 0.5) or 0.5) if not (0.05 <= scale <= 1.0): scale = 0.5 long_side = max(h, w) max_dim = max(64, int(long_side * scale)) if long_side > max_dim: det_scale = max_dim / long_side det_w = int(w * det_scale) det_h = int(h * det_scale) img_det = cv2.resize(img_bgr, (det_w, det_h), interpolation=cv2.INTER_LINEAR) inv_scale = 1.0 / det_scale size_range_det = ( max(4, int(getattr(config, "TRIANGLE_SIZE_RANGE", (8, 500))[0] * det_scale)), max(8, int(getattr(config, "TRIANGLE_SIZE_RANGE", (8, 500))[1] * det_scale)), ) else: img_det = img_bgr inv_scale = 1.0 size_range_det = getattr(config, "TRIANGLE_SIZE_RANGE", (8, 500)) gray = cv2.cvtColor(img_det, cv2.COLOR_BGR2GRAY) markers_det = detect_triangle_markers( gray, orig_gray=gray, size_range=size_range_det, verbose=bool(args.verbose), ) if inv_scale != 1.0 and markers_det: for m in markers_det: m["center"] = [m["center"][0] * inv_scale, m["center"][1] * inv_scale] m["corners"] = [[c[0] * inv_scale, c[1] * inv_scale] for c in m["corners"]] print("[tri-test] markers_found:", len(markers_det), "ids:", [m.get("id") for m in markers_det]) t0 = time.time() tri = try_triangle_scoring( img_rgb, # try_triangle_scoring 期望 RGB laser_point, pos, K, dist, size_range=getattr(config, "TRIANGLE_SIZE_RANGE", (8, 500)), ) dt_ms = int(round((time.time() - t0) * 1000)) print("[tri-test] elapsed_ms:", dt_ms) print(json.dumps(tri, ensure_ascii=False, indent=2)) if args.out: out_path = args.out # 允许传目录（如 ./），自动生成文件名；未带扩展名时默认 .jpg if out_path.endswith("/") or out_path.endswith("\\") or os.path.isdir(out_path): out_path = os.path.join(out_path, "tri_out.jpg") root, ext = os.path.splitext(out_path) if not ext: out_path = root + ".jpg" # 若 try_triangle_scoring 失败且没带回 markers，至少把候选 markers 画出来，方便肉眼判断 tri_for_draw = tri if isinstance(tri, dict) else {"ok": False} if not tri_for_draw.get("markers") and markers_det: tri_for_draw = dict(tri_for_draw) tri_for_draw["markers"] = markers_det out_img = _draw_tri_debug(img_bgr, tri_for_draw) ok = cv2.imwrite(out_path, out_img) if not ok: raise SystemExit(f"写图失败（可能是不支持的扩展名）：{out_path}") print("[tri-test] wrote:", out_path) if __name__ == "__main__": main()