From 824b9d77861ce4f612e4c343fda9c3bb40257748 Mon Sep 17 00:00:00 2001 From: Havoc <2993167370@qq.com> Date: Sat, 31 May 2025 10:21:55 +0800 Subject: [PATCH] =?UTF-8?q?=E5=A2=9E=E5=BC=BA=E5=8F=8C=E8=BD=A8=E8=BF=B9?= =?UTF-8?q?=E7=BA=BF=E6=A3=80=E6=B5=8B=E5=8A=9F=E8=83=BD=EF=BC=8C=E6=96=B0?= =?UTF-8?q?=E5=A2=9E=E5=9F=BA=E4=BA=8E=E4=B8=AD=E5=BF=83=E7=BA=BF=E7=9A=84?= =?UTF-8?q?=E6=A3=80=E6=B5=8B=E6=96=B9=E6=B3=95=EF=BC=8C=E4=BC=98=E5=8C=96?= =?UTF-8?q?=E4=BA=86=E5=8F=82=E6=95=B0=E8=AE=BE=E7=BD=AE=E5=92=8C=E8=AF=84?= =?UTF-8?q?=E5=88=86=E6=9C=BA=E5=88=B6=EF=BC=8C=E6=8F=90=E9=AB=98=E4=BA=86?= =?UTF-8?q?=E5=AF=B9=E4=B8=8D=E5=90=8C=E8=B7=AF=E5=86=B5=E7=9A=84=E5=A4=84?= =?UTF-8?q?=E7=90=86=E8=83=BD=E5=8A=9B=E5=92=8C=E6=A3=80=E6=B5=8B=E5=87=86?= =?UTF-8?q?=E7=A1=AE=E6=80=A7=E3=80=82=E5=90=8C=E6=97=B6=EF=BC=8C=E6=9B=B4?= =?UTF-8?q?=E6=96=B0=E4=BA=86=E6=97=A5=E5=BF=97=E8=AE=B0=E5=BD=95=E4=BB=A5?= =?UTF-8?q?=E4=BE=BF=E4=BA=8E=E8=B0=83=E8=AF=95=E5=92=8C=E7=BB=93=E6=9E=9C?= =?UTF-8?q?=E8=BF=BD=E8=B8=AA=E3=80=82?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- utils/detect_dual_track_lines.py | 1472 ++++++++++++++++++++++++++++++ 1 file changed, 1472 insertions(+) diff --git a/utils/detect_dual_track_lines.py b/utils/detect_dual_track_lines.py index 9197f8b..4d2723a 100644 --- a/utils/detect_dual_track_lines.py +++ b/utils/detect_dual_track_lines.py @@ -5,3 +5,1475 @@ import datetime from utils.log_helper import LogHelper, get_logger, section, info, debug, warning, error, success, timing +def detect_dual_track_lines(image, observe=False, delay=1000, save_log=True, stone_path_mode=False): + """ + 检测左右两条平行的黄色轨道线，优化后能够更准确处理各种路况 + + 参数: + image: 输入图像，可以是文件路径或者已加载的图像数组 + observe: 是否输出中间状态信息和可视化结果，默认为False + delay: 展示每个步骤的等待时间(毫秒) + save_log: 是否保存日志和图像 + stone_path_mode: 石板路模式，针对石板路上的黄线进行特殊处理 + + 返回: + tuple: (中心线信息, 左轨迹线信息, 右轨迹线信息) + 如果检测失败返回(None, None, None) + + 改进: + - 使用多点采样+多项式拟合计算更准确的中心线 + - 优化对左右轨道线的选择，考虑平行性和合理宽度 + - 增强对倾斜轨道和石板路的处理能力 + """ + # 如果输入是字符串（文件路径），则加载图像 + if isinstance(image, str): + img = cv2.imread(image) + else: + img = image.copy() + + if img is None: + error("无法加载图像", "失败") + return None, None, None + + # 获取图像尺寸 + height, width = img.shape[:2] + + # 计算图像中间区域的范围 + center_x = width // 2 + + # 转换到HSV颜色空间以便更容易提取黄色 + hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV) + + # 根据是否为石板路模式选择不同的参数 + if stone_path_mode: + # 石板路上的黄线通常对比度更低，需要更宽松的颜色范围 + lower_yellow = np.array([8, 50, 50]) # 更宽松的黄色下限 + upper_yellow = np.array([45, 255, 255]) # 更宽松的黄色上限 + else: + # 标准黄色的HSV范围 + lower_yellow = np.array([15, 80, 80]) + upper_yellow = np.array([35, 255, 255]) + + # 创建黄色的掩码 + mask = cv2.inRange(hsv, lower_yellow, upper_yellow) + + # 应用对比度增强 + if stone_path_mode: + # 在处理掩码前先对原图像进行预处理 + # 对原图进行自适应直方图均衡化增强对比度 + lab = cv2.cvtColor(img, cv2.COLOR_BGR2LAB) + l, a, b = cv2.split(lab) + clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8, 8)) + l = clahe.apply(l) + lab = cv2.merge((l, a, b)) + enhanced_img = cv2.cvtColor(lab, cv2.COLOR_LAB2BGR) + + # 用增强后的图像重新检测黄色 + enhanced_hsv = cv2.cvtColor(enhanced_img, cv2.COLOR_BGR2HSV) + enhanced_mask = cv2.inRange(enhanced_hsv, lower_yellow, upper_yellow) + + # 组合原始掩码和增强掩码 + mask = cv2.bitwise_or(mask, enhanced_mask) + + if observe: + debug("增强对比度和颜色检测", "处理") + cv2.imshow("增强对比度", enhanced_img) + cv2.imshow("增强后的黄色掩码", enhanced_mask) + cv2.waitKey(delay) + + # 形态学操作以改善掩码 + if stone_path_mode: + # 石板路上的线条可能更细更断续，使用更大的膨胀核和更多的迭代次数 + kernel = np.ones((7, 7), np.uint8) + mask = cv2.dilate(mask, kernel, iterations=2) + mask = cv2.erode(mask, np.ones((3, 3), np.uint8), iterations=1) + else: + kernel = np.ones((5, 5), np.uint8) + mask = cv2.dilate(mask, kernel, iterations=1) + mask = cv2.erode(mask, np.ones((3, 3), np.uint8), iterations=1) + + if observe: + debug("步骤1: 创建黄色掩码", "处理") + cv2.imshow("黄色掩码", mask) + cv2.waitKey(delay) + + # 裁剪底部区域重点关注近处的黄线 + bottom_roi_height = int(height * 0.6) # 增加关注区域到图像底部60% + bottom_roi = mask[height-bottom_roi_height:, :] + + if observe: + debug("步骤1.5: 底部区域掩码", "处理") + cv2.imshow("底部区域掩码", bottom_roi) + cv2.waitKey(delay) + + # 边缘检测 - 针对石板路调整参数 + if stone_path_mode: + edges = cv2.Canny(mask, 20, 100, apertureSize=3) # 进一步降低阈值以捕捉更弱的边缘 + else: + edges = cv2.Canny(mask, 50, 150, apertureSize=3) + + if observe: + debug("步骤2: 边缘检测", "处理") + cv2.imshow("边缘检测", edges) + cv2.waitKey(delay) + + # 霍夫变换检测直线 - 根据是否为石板路调整参数 + if stone_path_mode: + # 石板路上的线段可能更短更断续，使用更宽松的参数 + lines = cv2.HoughLinesP(edges, 1, np.pi/180, threshold=15, + minLineLength=width*0.02, maxLineGap=60) + else: + lines = cv2.HoughLinesP(edges, 1, np.pi/180, threshold=25, + minLineLength=width*0.05, maxLineGap=40) + + if lines is None or len(lines) == 0: + error("未检测到直线", "失败") + return None, None, None + + if observe: + debug(f"步骤3: 检测到 {len(lines)} 条直线", "处理") + lines_img = img.copy() + for line in lines: + x1, y1, x2, y2 = line[0] + cv2.line(lines_img, (x1, y1), (x2, y2), (0, 255, 0), 2) + cv2.imshow("检测到的直线", lines_img) + cv2.waitKey(delay) + + # 筛选近似垂直的线 + vertical_lines = [] + for line in lines: + x1, y1, x2, y2 = line[0] + + # 优先选择图像底部的线 + if y1 < height * 0.4 and y2 < height * 0.4: + continue # 忽略上半部分的线 + + # 计算斜率 (避免除零错误) + if abs(x2 - x1) < 5: # 几乎垂直的线 + slope = 100 # 设置一个较大的值表示接近垂直 + else: + slope = (y2 - y1) / (x2 - x1) + + # 根据是否为石板路调整斜率阈值 + min_slope_threshold = 0.4 if stone_path_mode else 0.75 + + # 筛选接近垂直的线 (斜率较大)，但允许更多倾斜度 + if abs(slope) > min_slope_threshold: + line_length = np.sqrt((x2-x1)**2 + (y2-y1)**2) + # 计算线的中点x坐标 + mid_x = (x1 + x2) / 2 + # 计算线的中点y坐标 + mid_y = (y1 + y2) / 2 + # 保存线段、其坐标、斜率和长度 + vertical_lines.append((line[0], mid_x, mid_y, slope, line_length)) + + if len(vertical_lines) < 2: + # 石板路模式下，尝试放宽斜率条件重新筛选线段 + if stone_path_mode: + vertical_lines = [] + for line in lines: + x1, y1, x2, y2 = line[0] + + # 仍然优先选择图像底部的线 + if y1 < height * 0.6 and y2 < height * 0.6: + continue # 忽略上部分的线 + + # 计算斜率 (避免除零错误) + if abs(x2 - x1) < 5: # 几乎垂直的线 + slope = 100 + else: + slope = (y2 - y1) / (x2 - x1) + + # 使用更宽松的斜率阈值 + if abs(slope) > 0.2: # 进一步放宽斜率阈值 + line_length = np.sqrt((x2-x1)**2 + (y2-y1)**2) + mid_x = (x1 + x2) / 2 + mid_y = (y1 + y2) / 2 + vertical_lines.append((line[0], mid_x, mid_y, slope, line_length)) + + if len(vertical_lines) < 2: + error("未检测到足够的垂直线", "失败") + return None, None, None + + if observe: + debug(f"步骤4: 找到 {len(vertical_lines)} 条垂直线", "处理") + v_lines_img = img.copy() + for line_info in vertical_lines: + line, _, _, slope, _ = line_info + x1, y1, x2, y2 = line + cv2.line(v_lines_img, (x1, y1), (x2, y2), (0, 255, 255), 2) + # 显示斜率 + cv2.putText(v_lines_img, f"{slope:.2f}", ((x1+x2)//2, (y1+y2)//2), + cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1) + cv2.imshow("垂直线", v_lines_img) + cv2.waitKey(delay) + + # 优先选择更接近图像底部的线 - 根据y坐标均值排序 + vertical_lines.sort(key=lambda x: x[2], reverse=True) # 按mid_y从大到小排序 + + # 石板路模式下，可能需要处理断续的线段或合并相近的线段 + if stone_path_mode and len(vertical_lines) >= 3: + # 尝试合并相近的线段 + merged_lines = [] + processed = [False] * len(vertical_lines) + + for i in range(len(vertical_lines)): + if processed[i]: + continue + + current_line = vertical_lines[i] + _, current_mid_x, current_mid_y, current_slope, current_length = current_line + + # 查找相近的线段 + similar_lines = [current_line] + processed[i] = True + + for j in range(i+1, len(vertical_lines)): + if processed[j]: + continue + + candidate_line = vertical_lines[j] + _, candidate_mid_x, candidate_mid_y, candidate_slope, candidate_length = candidate_line + + # 如果x坐标接近且斜率相似，认为是同一条线的不同部分 + x_diff = abs(current_mid_x - candidate_mid_x) + slope_diff = abs(current_slope - candidate_slope) + y_diff = abs(current_mid_y - candidate_mid_y) + + # 放宽相似线段的判断条件 + if ((x_diff < width * 0.08 and slope_diff < 0.4) or # 更宽松的x差异和斜率差异 + (x_diff < width * 0.05 and y_diff < height * 0.2)): # 或者x和y都比较接近 + similar_lines.append(candidate_line) + processed[j] = True + + # 如果找到多条相近的线，合并它们 + if len(similar_lines) > 1: + # 按线长度加权合并 + total_weight = sum(line[4] for line in similar_lines) + merged_x1 = sum(line[0][0] * line[4] for line in similar_lines) / total_weight + merged_y1 = sum(line[0][1] * line[4] for line in similar_lines) / total_weight + merged_x2 = sum(line[0][2] * line[4] for line in similar_lines) / total_weight + merged_y2 = sum(line[0][3] * line[4] for line in similar_lines) / total_weight + + merged_line = (np.array([int(merged_x1), int(merged_y1), + int(merged_x2), int(merged_y2)]), + (merged_x1 + merged_x2) / 2, + (merged_y1 + merged_y2) / 2, + (merged_y2 - merged_y1) / (merged_x2 - merged_x1 + 1e-6), + np.sqrt((merged_x2-merged_x1)**2 + (merged_y2-merged_y1)**2)) + + merged_lines.append(merged_line) + else: + merged_lines.append(current_line) + + vertical_lines = merged_lines + + if observe: + debug(f"步骤4.5: 合并后找到 {len(vertical_lines)} 条垂直线", "处理") + merged_img = img.copy() + for line_info in vertical_lines: + line, _, _, slope, _ = line_info + x1, y1, x2, y2 = line + cv2.line(merged_img, (int(x1), int(y1)), (int(x2), int(y2)), (255, 0, 255), 2) + # 显示斜率 + cv2.putText(merged_img, f"{slope:.2f}", (int((x1+x2)//2), int((y1+y2)//2)), + cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1) + cv2.imshow("合并后的垂直线", merged_img) + cv2.waitKey(delay) + + # 按x坐标将线分为左右两组 + left_lines = [line for line in vertical_lines if line[1] < center_x] + right_lines = [line for line in vertical_lines if line[1] > center_x] + + # 如果任一侧没有检测到线，则放宽左右两侧线的分组条件 + if not left_lines or not right_lines: + # 按x坐标排序所有垂直线 + vertical_lines.sort(key=lambda x: x[1]) + + # 如果有至少两条线，将最左侧的线作为左轨迹线，最右侧的线作为右轨迹线 + if len(vertical_lines) >= 2: + left_lines = [vertical_lines[0]] + right_lines = [vertical_lines[-1]] + else: + error("左侧或右侧未检测到轨迹线", "失败") + return None, None, None + + if observe: + debug(f"左侧候选线数量: {len(left_lines)}, 右侧候选线数量: {len(right_lines)}", "线候选") + + # 优化说明：在默认模式下，评分函数和线对选择都优先考虑更靠近图像中心的线段 + # 这有助于减少对图像边缘可能出现的干扰线的选择，提高轨道线检测的准确性 + + # 改进的评分函数 - 同时考虑斜率、位置、长度和在图像中的位置 + def score_line(line_info, is_left): + _, mid_x, mid_y, slope, length = line_info + line_points = line_info[0] # 获取线段的端点坐标 + x1, y1, x2, y2 = line_points + + # 确保线段从上到下排序 + if y1 > y2: + x1, x2 = x2, x1 + y1, y2 = y2, y1 + + # 线段靠近底部评分 - y越大（越靠近底部）分数越高 + bottom_ratio = mid_y / height + y_score = min(1.0, bottom_ratio * 1.2) # 适当提高底部线段的权重 + + # 线段长度评分 - 线越长分数越高 + # 调整长度评分，更倾向于较长的线段 + length_ratio = length / (height * 0.3) + length_score = min(1.0, length_ratio * 1.2) # 适当提高长线段的权重 + + # 位置评分 - 线段位于预期位置得分高 + # 根据是否为石板路模式调整预期位置 + center_x = width / 2 + if stone_path_mode: + # 石板路上的轨道宽度可能不同 + expected_track_width = width * 0.5 # 石板路轨道宽度估计 + else: + # 普通轨道的期望位置 - 默认模式下更靠近中心 + expected_track_width = width * 0.4 # 普通轨道宽度估计，更窄以接近中心 + + # 计算预期的线位置（基于图像中心和轨道宽度） + expected_x = center_x - expected_track_width * 0.5 if is_left else center_x + expected_track_width * 0.5 + + # 计算中点与预期位置的偏差 + x_distance = abs(mid_x - expected_x) + x_score = max(0, 1.0 - x_distance / (width * 0.25)) + + # 计算到图像中心的距离得分 - 默认模式下更重视靠近中心的线 + distance_to_center = abs(mid_x - center_x) + center_proximity_score = max(0, 1.0 - distance_to_center / (width * 0.4)) + + # 计算底部点与预期位置的偏差 + # 估计线延伸到底部的x坐标 + bottom_x = x1 + if abs(y2 - y1) > 1: # 非水平线 + t = (height - y1) / (y2 - y1) + if t >= 0: # 确保是向下延伸 + bottom_x = x1 + t * (x2 - x1) + + bottom_x_distance = abs(bottom_x - expected_x) + bottom_x_score = max(0, 1.0 - bottom_x_distance / (width * 0.25)) + + # 斜率评分 - 轨道线应该有一定的倾斜度 + # 判断是否几乎垂直 + if abs(slope) > 5: + # 几乎垂直的线，不太可能是轨道线 + slope_score = 0.3 # 给予低分但不完全排除 + else: + # 期望的斜率 - 左右轨道线应该稍有内倾 + # 左侧期望负斜率，右侧期望正斜率 + ideal_slope_magnitude = 0.8 # 适当倾斜 + ideal_slope = -ideal_slope_magnitude if is_left else ideal_slope_magnitude + + # 检查斜率符号是否正确 + if (is_left and slope > 0) or (not is_left and slope < 0): + # 斜率符号不对，大幅降低评分 + slope_sign_score = 0.3 + else: + slope_sign_score = 1.0 + + # 计算与理想斜率的接近度 + slope_diff = abs(slope - ideal_slope) + slope_magnitude_score = max(0, 1.0 - slope_diff / 2.0) + + # 综合斜率符号和幅度得分 + slope_score = slope_sign_score * slope_magnitude_score + + # 线段直线度评分 - 使用端点拟合的直线与原始线段的接近度 + # 在这里我们已经假设线段是直线，所以直线度是100% + + # 线段在图像中的位置评分 - 轨道线应该大致垂直并且从底部延伸 + # 检查线段是否从底部区域开始 + bottom_region_threshold = height * 0.7 # 底部30%区域 + reaches_bottom = max(y1, y2) > bottom_region_threshold + bottom_reach_score = 1.0 if reaches_bottom else 0.5 + + # 综合评分 - 调整权重 + if stone_path_mode: + # 石板路模式下更关注位置和底部接近程度 + final_score = ( + y_score * 0.25 + # 底部接近度 + length_score * 0.15 + # 线段长度 + x_score * 0.15 + # 中点位置 + bottom_x_score * 0.2 + # 底部点位置 + slope_score * 0.15 + # 斜率合适性 + bottom_reach_score * 0.1 # 是否到达底部 + ) + else: + # 普通轨道模式下更关注中心接近性 + final_score = ( + y_score * 0.15 + # 底部接近度 + length_score * 0.15 + # 线段长度 + x_score * 0.15 + # 中点位置 + center_proximity_score * 0.2 + # 与中心的接近度 (新增) + bottom_x_score * 0.15 + # 底部点位置 + slope_score * 0.1 + # 斜率合适性 + bottom_reach_score * 0.1 # 是否到达底部 + ) + + return final_score + + # 如果有多条线，评估左右线对是否平行 + if len(left_lines) > 0 and len(right_lines) > 0: + # 计算最佳的左右线对 + best_pair_score = -1 + best_left_line = None + best_right_line = None + + # 先对左右线按照评分排序，只考虑评分较高的候选线（减少计算量） + left_lines = sorted(left_lines, key=lambda line: score_line(line, True), reverse=True) + right_lines = sorted(right_lines, key=lambda line: score_line(line, False), reverse=True) + + # 只考虑前几名的线段 + max_candidates = min(5, len(left_lines), len(right_lines)) + left_candidates = left_lines[:max_candidates] + right_candidates = right_lines[:max_candidates] + + for left_line in left_candidates: + left_score = score_line(left_line, True) + left_slope = left_line[3] + + for right_line in right_candidates: + right_score = score_line(right_line, False) + right_slope = right_line[3] + + # 计算两条线的斜率差异 - 平行线斜率应该相近但符号相反 + # 对于接近垂直的线，使用符号相反但绝对值相近作为判断依据 + if abs(left_slope) > 5 and abs(right_slope) > 5: + # 几乎垂直的线，判断它们是否都几乎垂直 + slope_diff = abs(abs(left_slope) - abs(right_slope)) / max(abs(left_slope), abs(right_slope)) + parallel_score = max(0, 1.0 - slope_diff * 3.0) + else: + # 非垂直线，检查它们是否平行且方向相反(一个正斜率，一个负斜率) + if left_slope * right_slope < 0: # 一正一负 + slope_diff = abs(abs(left_slope) - abs(right_slope)) / max(0.1, max(abs(left_slope), abs(right_slope))) + parallel_score = max(0, 1.0 - slope_diff * 2.0) + else: + # 斜率符号相同，不太可能是轨道线对 + parallel_score = 0.0 + + # 计算两条线的宽度是否合理 + left_x = left_line[1] + right_x = right_line[1] + track_width = right_x - left_x + + # 检查宽度是否正常 + if track_width <= 0: + # 宽度为负，不合理 + width_score = 0.0 + else: + # 轨道宽度应该在合理范围内 - 调整范围更加精确 + if stone_path_mode: + expected_width = width * 0.5 # 石板路可能更宽一些 + allowed_deviation = width * 0.3 # 允许的偏差范围 + else: + expected_width = width * 0.4 # 普通轨道相对窄一些，更靠近中心 + allowed_deviation = width * 0.2 # 允许的偏差范围，更精确 + + width_diff = abs(track_width - expected_width) + width_score = max(0, 1.0 - width_diff / allowed_deviation) + + # 计算线段的垂直对称性 - 理想的轨道线应该大致以图像中心为对称轴 + center_x = width / 2 + left_dist_to_center = abs(center_x - left_x) + right_dist_to_center = abs(right_x - center_x) + + # 计算对称性得分，如果左右距离中心的距离相近，得分高 + symmetry_diff = abs(left_dist_to_center - right_dist_to_center) / max(left_dist_to_center, right_dist_to_center) + symmetry_score = max(0, 1.0 - symmetry_diff) + + # 底部点距离评分 - 确保两条线底部的点在合理距离内 + # 估计左右线段延伸到图像底部时的x坐标 + left_x1, left_y1, left_x2, left_y2 = left_line[0] + right_x1, right_y1, right_x2, right_y2 = right_line[0] + + left_bottom_x = left_x1 + if abs(left_y2 - left_y1) > 1: + t = (height - left_y1) / (left_y2 - left_y1) + left_bottom_x = left_x1 + t * (left_x2 - left_x1) + + right_bottom_x = right_x1 + if abs(right_y2 - right_y1) > 1: + t = (height - right_y1) / (right_y2 - right_y1) + right_bottom_x = right_x1 + t * (right_x2 - right_x1) + + bottom_width = right_bottom_x - left_bottom_x + + if bottom_width <= 0: + bottom_width_score = 0.0 + else: + bottom_width_diff = abs(bottom_width - expected_width) + bottom_width_score = max(0, 1.0 - bottom_width_diff / allowed_deviation) + + # 综合评分 - 调整权重 + # 更重视平行性和底部宽度 + if stone_path_mode: + pair_score = (left_score + right_score) * 0.3 + parallel_score * 0.25 + width_score * 0.2 + symmetry_score * 0.1 + bottom_width_score * 0.15 + else: + # 默认模式下，增加对中心接近性的权重 + # 计算左右线到中心的距离 + left_to_center = abs(left_x - center_x) + right_to_center = abs(right_x - center_x) + + # 标准化距离（与图像宽度相关） + left_center_ratio = left_to_center / (width * 0.5) + right_center_ratio = right_to_center / (width * 0.5) + + # 接近度得分 - 越接近中心分数越高 + left_center_score = max(0, 1.0 - left_center_ratio) + right_center_score = max(0, 1.0 - right_center_ratio) + center_proximity_score = (left_center_score + right_center_score) / 2 + + # 给予中心接近性更高的权重 + pair_score = (left_score + right_score) * 0.25 + parallel_score * 0.2 + width_score * 0.15 + symmetry_score * 0.1 + bottom_width_score * 0.1 + center_proximity_score * 0.2 + + if pair_score > best_pair_score: + best_pair_score = pair_score + best_left_line = left_line + best_right_line = right_line + + # 如果找到了最佳对，则使用它们 + if best_left_line is not None and best_right_line is not None and best_pair_score > 0.4: # 要求最低评分 + left_line = best_left_line + right_line = best_right_line + + if observe: + debug(f"选择最佳线对，评分: {best_pair_score:.2f}", "线对") + else: + # 如果未找到合适的线对，则使用各自评分最高的线 + left_lines = sorted(left_lines, key=lambda line: score_line(line, True), reverse=True) + right_lines = sorted(right_lines, key=lambda line: score_line(line, False), reverse=True) + left_line = left_lines[0] + right_line = right_lines[0] + + if observe: + debug("未找到合适的线对，使用各自评分最高的线", "线对") + else: + # 如果只有单侧有线，使用评分最高的线 + left_lines = sorted(left_lines, key=lambda line: score_line(line, True), reverse=True) + right_lines = sorted(right_lines, key=lambda line: score_line(line, False), reverse=True) + left_line = left_lines[0] if left_lines else None + right_line = right_lines[0] if right_lines else None + + # 确保两侧都有线 + if left_line is None or right_line is None: + error("无法确定合适的左右轨迹线对", "失败") + return None, None, None + + # 获取两条线的坐标 + left_x1, left_y1, left_x2, left_y2 = left_line[0] + right_x1, right_y1, right_x2, right_y2 = right_line[0] + + # 确保线段的顺序是从上到下 + if left_y1 > left_y2: + left_x1, left_x2 = left_x2, left_x1 + left_y1, left_y2 = left_y2, left_y1 + + if right_y1 > right_y2: + right_x1, right_x2 = right_x2, right_x1 + right_y1, right_y2 = right_y2, right_y1 + + # 尝试延长线段到图像底部，处理被石板路部分遮挡的情况 + left_extended_y2 = height + if abs(left_x2 - left_x1) < 5: # 几乎垂直 + left_extended_x2 = left_x2 + else: + left_slope = (left_y2 - left_y1) / (left_x2 - left_x1) + left_extended_x2 = left_x1 + (left_extended_y2 - left_y1) / left_slope + + right_extended_y2 = height + if abs(right_x2 - right_x1) < 5: # 几乎垂直 + right_extended_x2 = right_x2 + else: + right_slope = (right_y2 - right_y1) / (right_x2 - right_x1) + right_extended_x2 = right_x1 + (right_extended_y2 - right_y1) / right_slope + + # 更新线段端点为延长后的坐标 + left_x2, left_y2 = int(left_extended_x2), left_extended_y2 + right_x2, right_y2 = int(right_extended_x2), right_extended_y2 + + # 改进的中心线计算方法 + # 首先确定两条轨迹线的有效部分 - 以两条线段的y坐标重叠部分为准 + min_y = max(left_y1, right_y1) + max_y = min(left_y2, right_y2) + + # 如果两条线没有重叠的y部分，则使用整个图像范围 + if min_y >= max_y: + min_y = 0 + max_y = height + + # 计算中间路径点的方式：在多个高度上计算左右线的中点，然后拟合中心线 + num_points = 20 # 增加采样点数量以提高精度 + center_points = [] + + # 优化采样策略 - 在底部区域采样更密集 + sample_ys = [] + + # 前半部分采样点均匀分布 + first_half = np.linspace(min_y, min_y + (max_y - min_y) * 0.5, num_points // 4) + # 后半部分(更靠近底部)采样点更密集 + second_half = np.linspace(min_y + (max_y - min_y) * 0.5, max_y, num_points - num_points // 4) + + sample_ys = np.concatenate([first_half, second_half]) + + for y in sample_ys: + # 计算左侧线在当前y值处的x坐标 + if abs(left_y2 - left_y1) < 1: # 防止除零 + left_x = left_x1 + else: + t = (y - left_y1) / (left_y2 - left_y1) + left_x = left_x1 + t * (left_x2 - left_x1) + + # 计算右侧线在当前y值处的x坐标 + if abs(right_y2 - right_y1) < 1: # 防止除零 + right_x = right_x1 + else: + t = (y - right_y1) / (right_y2 - right_y1) + right_x = right_x1 + t * (right_x2 - right_x1) + + # 计算中心点 + center_x = (left_x + right_x) / 2 + center_points.append((center_x, y)) + + # 使用采样的中心点拟合中心线 + center_xs = np.array([p[0] for p in center_points]) + center_ys = np.array([p[1] for p in center_points]) + + # 使用多项式拟合改进中心线 - 选择合适的多项式阶数 + if len(center_points) >= 5: # 需要更多点来拟合更高阶多项式 + try: + # 尝试不同阶数的多项式拟合，选择最佳结果 + best_poly_coeffs = None + best_error = float('inf') + + for degree in [1, 2, 3]: # 尝试1-3阶多项式 + try: + # 使用多项式拟合 + poly_coeffs = np.polyfit(center_ys, center_xs, degree) + poly_func = np.poly1d(poly_coeffs) + + # 计算拟合误差 + predicted_xs = poly_func(center_ys) + error = np.mean(np.abs(predicted_xs - center_xs)) + + # 如果误差更小，更新最佳拟合 + if error < best_error: + best_error = error + best_poly_coeffs = poly_coeffs + except: + continue + + # 如果找到了有效的拟合，使用它 + if best_poly_coeffs is not None: + poly_coeffs = best_poly_coeffs + poly_func = np.poly1d(poly_coeffs) + + # 为了提高底部拟合精度，给予底部区域更高的权重重新拟合 + weights = np.ones_like(center_ys) + + # 计算距离底部的归一化距离 (0表示底部，1表示顶部) + normalized_distance = (max_y - center_ys) / max(1, (max_y - min_y)) + + # 设置权重，底部权重更高 + weights = 1.0 + 2.0 * (1.0 - normalized_distance) + + # 使用加权多项式拟合 + try: + weighted_poly_coeffs = np.polyfit(center_ys, center_xs, len(best_poly_coeffs) - 1, w=weights) + weighted_poly_func = np.poly1d(weighted_poly_coeffs) + + # 比较加权拟合与原始拟合 + weighted_predicted_xs = weighted_poly_func(center_ys) + weighted_error = np.mean(np.abs(weighted_predicted_xs - center_xs)) + + # 如果加权拟合更好，则使用它 + if weighted_error < best_error * 1.2: # 允许一定的误差增加，因为我们更关注底部精度 + poly_coeffs = weighted_poly_coeffs + poly_func = weighted_poly_func + except: + pass # 如果加权拟合失败，继续使用未加权的拟合 + + # 使用多项式生成中心线的起点和终点 + center_line_y1 = min_y + center_line_y2 = height # 延伸到图像底部 + + # 计算多项式在这些y值处的x坐标 + center_line_x1 = poly_func(center_line_y1) + center_line_x2 = poly_func(center_line_y2) + + # 计算中心线在图像底部的x坐标 - 用于计算偏离度 + bottom_x = poly_func(height) + + # 确保坐标在图像范围内 + bottom_x = max(0, min(width - 1, bottom_x)) + center_point = (int(bottom_x), int(height)) + + # 计算中心线的加权平均斜率 - 更关注底部区域的斜率 + if abs(center_line_y2 - center_line_y1) < 1: # 防止除零 + center_slope = 0 + else: + # 计算全局斜率 + global_slope = (center_line_x2 - center_line_x1) / (center_line_y2 - center_line_y1) + + # 计算底部区域的斜率，使用多个点来提高精度 + bottom_slopes = [] + bottom_region_start = height - height * 0.3 # 底部30%区域 + + if bottom_region_start > min_y: + # 在底部区域采样多个点计算局部斜率 + bottom_sample_count = 5 + bottom_ys = np.linspace(bottom_region_start, height, bottom_sample_count) + + for i in range(len(bottom_ys) - 1): + y1, y2 = bottom_ys[i], bottom_ys[i+1] + x1, x2 = poly_func(y1), poly_func(y2) + + # 确保坐标有效 + x1 = max(0, min(width - 1, x1)) + x2 = max(0, min(width - 1, x2)) + + local_slope = (x2 - x1) / max(0.1, (y2 - y1)) + bottom_slopes.append(local_slope) + + # 计算底部斜率的加权平均值，越靠近底部权重越高 + if bottom_slopes: + weights = np.linspace(1, 2, len(bottom_slopes)) + bottom_slope = np.average(bottom_slopes, weights=weights) + + # 加权平均，底部斜率权重更高 + center_slope = bottom_slope * 0.8 + global_slope * 0.2 + else: + center_slope = global_slope + else: + center_slope = global_slope + else: + # 如果所有多项式拟合都失败，退回到简单的线性拟合 + raise Exception("多项式拟合失败") + + except Exception as e: + warning(f"多项式拟合失败，使用简单中点计算: {e}", "拟合") + # 如果多项式拟合失败，退回到简单的中点计算方法 + center_line_x1 = (left_x1 + right_x1) / 2 + center_line_y1 = (left_y1 + right_y1) / 2 + center_line_x2 = (left_x2 + right_x2) / 2 + center_line_y2 = (left_y2 + right_y2) / 2 + + # 计算中心线的斜率 + if abs(center_line_x2 - center_line_x1) < 5: + center_slope = 100 # 几乎垂直 + else: + center_slope = (center_line_y2 - center_line_y1) / (center_line_x2 - center_line_x1) + + # 计算中心线延伸到图像底部的点 + if abs(center_slope) < 0.01: # 几乎水平 + bottom_x = center_line_x1 + else: + bottom_x = center_line_x1 + (height - center_line_y1) / center_slope + + bottom_x = max(0, min(width - 1, bottom_x)) + center_point = (int(bottom_x), int(height)) + else: + # 如果点数不足，退回到简单的中点计算方法 + center_line_x1 = (left_x1 + right_x1) / 2 + center_line_y1 = (left_y1 + right_y1) / 2 + center_line_x2 = (left_x2 + right_x2) / 2 + center_line_y2 = (left_y2 + right_y2) / 2 + + # 计算中心线的斜率 + if abs(center_line_x2 - center_line_x1) < 5: + center_slope = 100 # 几乎垂直 + else: + center_slope = (center_line_y2 - center_line_y1) / (center_line_x2 - center_line_x1) + + # 计算中心线延伸到图像底部的点 + if abs(center_slope) < 0.01: # 几乎水平 + bottom_x = center_line_x1 + else: + bottom_x = center_line_x1 + (height - center_line_y1) / center_slope + + bottom_x = max(0, min(width - 1, bottom_x)) + center_point = (int(bottom_x), int(height)) + + # 计算中心线与图像中心线的偏差 + deviation = bottom_x - center_x + + result_img = None + if observe or save_log: + result_img = img.copy() + # 绘制左右轨迹线 + try: + # 确保坐标在图像范围内并且是整数 + left_x1_safe = max(0, min(width - 1, left_x1)) + left_y1_safe = max(0, min(height - 1, left_y1)) + left_x2_safe = max(0, min(width - 1, left_x2)) + left_y2_safe = max(0, min(height - 1, left_y2)) + + cv2.line(result_img, (int(left_x1_safe), int(left_y1_safe)), + (int(left_x2_safe), int(left_y2_safe)), (255, 0, 0), 2) + except Exception as e: + warning(f"绘制左轨迹线错误: {e}", "绘图") + + try: + # 确保坐标在图像范围内并且是整数 + right_x1_safe = max(0, min(width - 1, right_x1)) + right_y1_safe = max(0, min(height - 1, right_y1)) + right_x2_safe = max(0, min(width - 1, right_x2)) + right_y2_safe = max(0, min(height - 1, right_y2)) + + cv2.line(result_img, (int(right_x1_safe), int(right_y1_safe)), + (int(right_x2_safe), int(right_y2_safe)), (0, 0, 255), 2) + except Exception as e: + warning(f"绘制右轨迹线错误: {e}", "绘图") + + # 绘制中心线 - 如果有多项式拟合，绘制拟合曲线 + if len(center_points) >= 3: + # 绘制拟合的中心曲线 + try: + curve_ys = np.linspace(min_y, height, 100) + curve_xs = poly_func(curve_ys) + + for i in range(len(curve_ys) - 1): + try: + # 确保坐标在图像范围内并且是整数 + x1 = max(0, min(width - 1, curve_xs[i])) + y1 = max(0, min(height - 1, curve_ys[i])) + x2 = max(0, min(width - 1, curve_xs[i+1])) + y2 = max(0, min(height - 1, curve_ys[i+1])) + + pt1 = (int(x1), int(y1)) + pt2 = (int(x2), int(y2)) + cv2.line(result_img, pt1, pt2, (0, 255, 0), 2) + except Exception as e: + warning(f"绘制曲线段错误: {e}", "绘图") + continue + + # 绘制采样点 + for pt in center_points: + try: + # 确保坐标在图像范围内并且是整数 + x = max(0, min(width - 1, pt[0])) + y = max(0, min(height - 1, pt[1])) + cv2.circle(result_img, (int(x), int(y)), 3, (0, 255, 255), -1) + except Exception as e: + warning(f"绘制采样点错误: {e}", "绘图") + continue + except Exception as e: + warning(f"绘制曲线错误: {e}", "绘图") + else: + # 绘制简单中心线 + try: + # 确保坐标在图像范围内并且是整数 + x1 = max(0, min(width - 1, center_line_x1)) + y1 = max(0, min(height - 1, center_line_y1)) + x2 = max(0, min(width - 1, center_line_x2)) + y2 = max(0, min(height - 1, center_line_y2)) + + cv2.line(result_img, (int(x1), int(y1)), (int(x2), int(y2)), (0, 255, 0), 2) + except Exception as e: + warning(f"绘制中心线错误: {e}", "绘图") + + # 绘制图像中心线 + try: + center_x_safe = max(0, min(width - 1, center_x)) + cv2.line(result_img, (int(center_x_safe), 0), (int(center_x_safe), height), (0, 0, 255), 1) + except Exception as e: + warning(f"绘制图像中心线错误: {e}", "绘图") + + # 标记中心点 + try: + # 确保中心点坐标在图像范围内 + center_x_safe = max(0, min(width - 1, center_point[0])) + center_y_safe = max(0, min(height - 1, center_point[1])) + cv2.circle(result_img, (int(center_x_safe), int(center_y_safe)), 10, (255, 0, 255), -1) + except Exception as e: + warning(f"绘制中心点错误: {e}", "绘图") + + # 显示偏差信息 + cv2.putText(result_img, f"Deviation: {deviation:.1f}px", (10, 30), + cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2) + + # 显示是否为石板路模式 + if stone_path_mode: + cv2.putText(result_img, "Stone Path Mode", (10, 60), + cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 255), 2) + + if observe: + cv2.imshow("轨迹线检测结果", result_img) + cv2.waitKey(delay) + + # 保存日志图像 + if save_log and result_img is not None: + timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S_%f") + log_dir = "logs/image" + os.makedirs(log_dir, exist_ok=True) + + # 保存结果图像 + result_img_path = os.path.join(log_dir, f"dual_track_{timestamp}.jpg") + cv2.imwrite(result_img_path, result_img) + + # 保存原图 + orig_img_path = os.path.join(log_dir, f"dual_track_orig_{timestamp}.jpg") + cv2.imwrite(orig_img_path, img) + + info(f"保存双轨迹线检测结果图像到: {result_img_path}", "日志") + info(f"保存原始图像到: {orig_img_path}", "日志") + + # 保存文本日志信息 + log_info = { + "timestamp": timestamp, + "center_point": (int(center_point[0]), int(center_point[1])), + "deviation": float(deviation), + "left_track_mid_x": float(left_line[1]), + "right_track_mid_x": float(right_line[1]), + "track_width": float(right_line[1] - left_line[1]), + "center_slope": float(center_slope), + "stone_path_mode": stone_path_mode + } + info(f"双轨迹线检测结果: {log_info}", "日志") + + # 创建左右轨迹线和中心线信息 + left_track_info = { + "line": (left_x1, left_y1, left_x2, left_y2), + "slope": left_line[3], + "x_mid": left_line[1] + } + + right_track_info = { + "line": (right_x1, right_y1, right_x2, right_y2), + "slope": right_line[3], + "x_mid": right_line[1] + } + + center_info = { + "point": (int(center_point[0]), int(center_point[1])), + "deviation": float(deviation), + "slope": float(center_slope), + "is_vertical": abs(center_slope) > 5.0, # 判断是否接近垂直 + "track_width": float(right_line[1] - left_line[1]), # 两轨迹线之间的距离 + "stone_path_mode": stone_path_mode # 记录是否使用了石板路模式 + } + + return center_info, left_track_info, right_track_info + +def detect_center_based_dual_track_lines(image, observe=False, delay=1000, save_log=True, expected_track_width_ratio=0.45): + """ + 通过先检测中心线，然后向两侧扩展来检测双轨迹线 + + 参数: + image: 输入图像，可以是文件路径或者已加载的图像数组 + observe: 是否输出中间状态信息和可视化结果，默认为False + delay: 展示每个步骤的等待时间(毫秒) + save_log: 是否保存日志和图像 + expected_track_width_ratio: 预期轨道宽度占图像宽度的比例，默认为0.45 + + 返回: + tuple: (中心线信息, 左轨迹线信息, 右轨迹线信息) + 如果检测失败返回(None, None, None) + """ + # 如果输入是字符串（文件路径），则加载图像 + if isinstance(image, str): + img = cv2.imread(image) + else: + img = image.copy() + + if img is None: + error("无法加载图像", "失败") + return None, None, None + + # 获取图像尺寸 + height, width = img.shape[:2] + + # 计算图像中间区域的范围 + center_x = width // 2 + + # 转换到HSV颜色空间以便更容易提取黄色 + hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV) + + # 标准黄色的HSV范围 + lower_yellow = np.array([15, 80, 80]) + upper_yellow = np.array([35, 255, 255]) + + # 创建黄色的掩码 + mask = cv2.inRange(hsv, lower_yellow, upper_yellow) + + # 对较暗的黄色使用更宽松的阈值 + lower_dark_yellow = np.array([10, 60, 60]) + upper_dark_yellow = np.array([40, 255, 255]) + dark_mask = cv2.inRange(hsv, lower_dark_yellow, upper_dark_yellow) + + # 合并掩码 + combined_mask = cv2.bitwise_or(mask, dark_mask) + + # 形态学操作以改善掩码 + kernel = np.ones((5, 5), np.uint8) + combined_mask = cv2.dilate(combined_mask, kernel, iterations=1) + combined_mask = cv2.erode(combined_mask, np.ones((3, 3), np.uint8), iterations=1) + + if observe: + debug("步骤1: 创建黄色掩码", "处理") + cv2.imshow("黄色掩码", combined_mask) + cv2.waitKey(delay) + + # 裁剪底部区域重点关注近处的黄线 + bottom_roi_height = int(height * 0.6) # 关注图像底部60% + bottom_roi = combined_mask[height-bottom_roi_height:, :] + + if observe: + debug("步骤1.5: 底部区域掩码", "处理") + cv2.imshow("底部区域掩码", bottom_roi) + cv2.waitKey(delay) + + # 边缘检测 + edges = cv2.Canny(combined_mask, 50, 150, apertureSize=3) + + if observe: + debug("步骤2: 边缘检测", "处理") + cv2.imshow("边缘检测", edges) + cv2.waitKey(delay) + + # 霍夫变换检测直线 + lines = cv2.HoughLinesP(edges, 1, np.pi/180, threshold=25, + minLineLength=width*0.05, maxLineGap=40) + + if lines is None or len(lines) == 0: + error("未检测到直线", "失败") + return None, None, None + + if observe: + debug(f"步骤3: 检测到 {len(lines)} 条直线", "处理") + lines_img = img.copy() + for line in lines: + x1, y1, x2, y2 = line[0] + cv2.line(lines_img, (x1, y1), (x2, y2), (0, 255, 0), 2) + cv2.imshow("检测到的直线", lines_img) + cv2.waitKey(delay) + + # 筛选近似垂直的线 + vertical_lines = [] + for line in lines: + x1, y1, x2, y2 = line[0] + + # 优先选择图像底部的线 + if y1 < height * 0.4 and y2 < height * 0.4: + continue # 忽略上半部分的线 + + # 计算斜率 (避免除零错误) + if abs(x2 - x1) < 5: # 几乎垂直的线 + slope = 100 # 设置一个较大的值表示接近垂直 + else: + slope = (y2 - y1) / (x2 - x1) + + # 筛选接近垂直的线 (斜率较大) + if abs(slope) > 0.5: # 降低斜率阈值以捕获更多候选线 + line_length = np.sqrt((x2-x1)**2 + (y2-y1)**2) + # 计算线的中点x坐标 + mid_x = (x1 + x2) / 2 + # 计算线的中点y坐标 + mid_y = (y1 + y2) / 2 + # 保存线段、其坐标、斜率和长度 + vertical_lines.append((line[0], mid_x, mid_y, slope, line_length)) + + if len(vertical_lines) < 1: + error("未检测到足够的垂直线", "失败") + return None, None, None + + if observe: + debug(f"步骤4: 找到 {len(vertical_lines)} 条垂直线", "处理") + v_lines_img = img.copy() + for line_info in vertical_lines: + line, _, _, slope, _ = line_info + x1, y1, x2, y2 = line + cv2.line(v_lines_img, (x1, y1), (x2, y2), (0, 255, 255), 2) + # 显示斜率 + cv2.putText(v_lines_img, f"{slope:.2f}", ((x1+x2)//2, (y1+y2)//2), + cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1) + cv2.imshow("垂直线", v_lines_img) + cv2.waitKey(delay) + + # 评分函数 - 用于找到最可能的中心线 + def score_center_line(line_info): + _, mid_x, mid_y, slope, length = line_info + + # 中心接近度 - 线越接近图像中心得分越高 + center_dist = abs(mid_x - center_x) + center_score = max(0, 1.0 - center_dist / (width * 0.25)) + + # 底部接近度 - y越大（越靠近底部）分数越高 + bottom_ratio = mid_y / height + y_score = min(1.0, bottom_ratio * 1.2) + + # 线段长度评分 - 线越长分数越高 + length_ratio = length / (height * 0.3) + length_score = min(1.0, length_ratio * 1.2) + + # 斜率评分 - 线应该接近垂直但不完全垂直 + if abs(slope) > 10: # 几乎垂直 + slope_score = 0.7 + else: + # 理想斜率在1-5之间 + ideal_slope = 2.0 + slope_diff = abs(abs(slope) - ideal_slope) + slope_score = max(0, 1.0 - slope_diff / 3.0) + + # 综合评分，重点是中心接近度 + final_score = ( + center_score * 0.5 + # 中心接近度权重最高 + y_score * 0.2 + # 底部接近度 + length_score * 0.2 + # 线段长度 + slope_score * 0.1 # 斜率合适性 + ) + + return final_score + + # 找到可能的中心线 + center_candidates = sorted(vertical_lines, key=score_center_line, reverse=True) + + # 如果有足够的候选线，取前3个进行评估 + if len(center_candidates) >= 3: + center_candidates = center_candidates[:3] + + # 获取左侧和右侧的线 + left_lines = [line for line in vertical_lines if line[1] < center_x - width * 0.1] # 确保在中心左侧一定距离 + right_lines = [line for line in vertical_lines if line[1] > center_x + width * 0.1] # 确保在中心右侧一定距离 + + # 按照到图像中心的距离排序 + left_lines.sort(key=lambda x: center_x - x[1]) # 从最靠近中心到最远 + right_lines.sort(key=lambda x: x[1] - center_x) # 从最靠近中心到最远 + + # 如果左侧或右侧没有足够的线，可能无法进行检测 + if not left_lines or not right_lines: + warning("左侧或右侧未检测到足够的线段", "检测") + # 尝试放宽条件 + left_lines = [line for line in vertical_lines if line[1] < center_x] + right_lines = [line for line in vertical_lines if line[1] > center_x] + + # 再次检查 + if not left_lines or not right_lines: + error("即使放宽条件，仍无法找到左右两侧的线段", "失败") + return None, None, None + + # 对于每个可能的中心线，尝试找到最佳的左右轨迹线对 + best_result = None + best_score = -1 + + for center_candidate in center_candidates: + center_line_coords = center_candidate[0] + center_x1, center_y1, center_x2, center_y2 = center_line_coords + center_mid_x = center_candidate[1] + + # 确保中心线从上到下排序 + if center_y1 > center_y2: + center_x1, center_x2 = center_x2, center_x1 + center_y1, center_y2 = center_y2, center_y1 + + # 扩展中心线到图像底部 + if abs(center_x2 - center_x1) < 5: # 几乎垂直 + center_extended_x2 = center_x2 + center_slope = 100 + else: + center_slope = (center_y2 - center_y1) / (center_x2 - center_x1) + center_extended_x2 = center_x1 + (height - center_y1) / center_slope + + center_x2, center_y2 = int(center_extended_x2), height + + # 计算预期的轨道宽度 + # 根据图像尺寸调整预期轨道宽度 + if width <= 640: # 小尺寸图像 + expected_track_width = width * expected_track_width_ratio + else: # 大尺寸图像 + expected_track_width = width * expected_track_width_ratio * 1.2 # 大图像可能需要更宽的轨道 + + half_track_width = expected_track_width / 2 + + # 基于中心线位置，预测左右轨道线的位置 + if abs(center_slope) > 5: # 几乎垂直中心线 + expected_left_x = center_mid_x - half_track_width + expected_right_x = center_mid_x + half_track_width + else: + # 考虑中心线的斜率计算预期位置 + perpendicular_slope = -1 / center_slope + len_perpendicular = np.sqrt(1 + perpendicular_slope**2) + unit_perp_x = 1 / len_perpendicular + + expected_left_x = center_mid_x - half_track_width * unit_perp_x + expected_right_x = center_mid_x + half_track_width * unit_perp_x + + # 找到最接近预期位置的左右轨道线 + # 给定一个预期位置和候选线，计算分数 + def score_track_line(line_info, expected_x, side): + _, mid_x, mid_y, slope, length = line_info + + # 位置接近度 + pos_dist = abs(mid_x - expected_x) + pos_score = max(0, 1.0 - pos_dist / (width * 0.2)) + + # 底部接近度 + bottom_ratio = mid_y / height + y_score = min(1.0, bottom_ratio * 1.2) + + # 线段长度 + length_ratio = length / (height * 0.3) + length_score = min(1.0, length_ratio * 1.2) + + # 斜率评分 - 检查是否与中心线斜率一致 + # 左侧线应该与中心线斜率相似或稍微向内倾斜 + # 右侧线应该与中心线斜率相似或稍微向内倾斜 + if abs(center_slope) > 5: # 中心线几乎垂直 + ideal_slope = 100 if abs(slope) > 5 else 5 * (1 if side == 'right' else -1) + else: + # 中心线不垂直，左右线应与中心线有相似但略有内倾的斜率 + sign_mult = 1 if side == 'right' else -1 + ideal_slope = center_slope + sign_mult * 0.2 + + slope_diff = abs(slope - ideal_slope) + slope_score = max(0, 1.0 - slope_diff / max(1, abs(ideal_slope))) + + # 针对左右侧的特殊评分条件 + if side == 'left': + # 左侧线的x坐标应该小于中心线 + if mid_x >= center_mid_x: + return 0.0 + # 并且应该在一个合理范围内（不要太远） + if mid_x < center_mid_x - width * 0.4: + return 0.0 + else: # 右侧 + # 右侧线的x坐标应该大于中心线 + if mid_x <= center_mid_x: + return 0.0 + # 并且应该在一个合理范围内（不要太远） + if mid_x > center_mid_x + width * 0.4: + return 0.0 + + # 综合评分，位置接近度最重要 + final_score = ( + pos_score * 0.5 + # 位置接近度 + y_score * 0.2 + # 底部接近度 + length_score * 0.2 + # 线长 + slope_score * 0.1 # 斜率合适性 + ) + + return final_score + + # 为左右轨道线评分 + left_scores = [(line, score_track_line(line, expected_left_x, 'left')) for line in left_lines] + right_scores = [(line, score_track_line(line, expected_right_x, 'right')) for line in right_lines] + + # 过滤掉评分为0的线 + left_scores = [item for item in left_scores if item[1] > 0] + right_scores = [item for item in right_scores if item[1] > 0] + + # 如果没有足够的候选线，跳过这个中心线 + if not left_scores or not right_scores: + continue + + # 按评分排序 + left_scores.sort(key=lambda x: x[1], reverse=True) + right_scores.sort(key=lambda x: x[1], reverse=True) + + # 取最高分的左右线 + best_left_line = left_scores[0][0] + best_right_line = right_scores[0][0] + + # 检查左右线之间的距离是否合理 + left_mid_x = best_left_line[1] + right_mid_x = best_right_line[1] + track_width = right_mid_x - left_mid_x + + # 计算轨道宽度评分 + width_ratio = track_width / expected_track_width + if 0.7 <= width_ratio <= 1.3: # 允许一定范围的偏差 + width_score = 1.0 + else: + width_score = max(0, 1.0 - abs(1.0 - width_ratio) * 2) + + # 计算左右线的对称性 + symmetry_score = 1.0 - abs((center_mid_x - left_mid_x) - (right_mid_x - center_mid_x)) / (track_width / 2) + symmetry_score = max(0, symmetry_score) + + # 总体评分 + pair_score = ( + left_scores[0][1] * 0.3 + # 左线评分 + right_scores[0][1] * 0.3 + # 右线评分 + width_score * 0.3 + # 轨道宽度合理性 + symmetry_score * 0.1 # 对称性 + ) + + if pair_score > best_score: + best_score = pair_score + best_result = (center_candidate, best_left_line, best_right_line, center_slope) + + # 如果没有找到合适的结果，返回失败 + if best_result is None: + error("未找到合适的左右轨迹线对", "失败") + return None, None, None + + # 提取最佳结果 + center_line, left_line, right_line, center_slope = best_result + + # 获取线段坐标 + center_coords = center_line[0] + left_coords = left_line[0] + right_coords = right_line[0] + + center_x1, center_y1, center_x2, center_y2 = center_coords + left_x1, left_y1, left_x2, left_y2 = left_coords + right_x1, right_y1, right_x2, right_y2 = right_coords + + # 确保线段从上到下排序 + if center_y1 > center_y2: + center_x1, center_x2 = center_x2, center_x1 + center_y1, center_y2 = center_y2, center_y1 + + if left_y1 > left_y2: + left_x1, left_x2 = left_x2, left_x1 + left_y1, left_y2 = left_y2, left_y1 + + if right_y1 > right_y2: + right_x1, right_x2 = right_x2, right_x1 + right_y1, right_y2 = right_y2, right_y1 + + # 计算中心线在图像底部的位置和偏离中心的距离 + if abs(center_x2 - center_x1) < 5: # 几乎垂直 + bottom_center_x = center_x2 + else: + center_slope = (center_y2 - center_y1) / (center_x2 - center_x1) + bottom_center_x = center_x1 + (height - center_y1) / center_slope + + # 确保坐标在图像范围内 + bottom_center_x = max(0, min(width - 1, bottom_center_x)) + deviation = bottom_center_x - center_x + + # 创建结果图像 + result_img = None + if observe or save_log: + result_img = img.copy() + + # 绘制中心线 + cv2.line(result_img, (center_x1, center_y1), (center_x2, center_y2), (0, 255, 0), 2) + + # 绘制左轨迹线 + cv2.line(result_img, (int(left_x1), int(left_y1)), (int(left_x2), int(left_y2)), (255, 0, 0), 2) + + # 绘制右轨迹线 + cv2.line(result_img, (int(right_x1), int(right_y1)), (int(right_x2), int(right_y2)), (0, 0, 255), 2) + + # 绘制图像中心线 + cv2.line(result_img, (center_x, 0), (center_x, height), (0, 0, 255), 1) + + # 标记中心点 + cv2.circle(result_img, (int(bottom_center_x), height), 10, (255, 0, 255), -1) + + # 显示偏差信息 + cv2.putText(result_img, f"Deviation: {deviation:.1f}px", (10, 30), + cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2) + + if observe: + cv2.imshow("改进的中心线检测结果", result_img) + cv2.waitKey(delay) + + # 保存日志图像 + if save_log and result_img is not None: + timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S_%f") + log_dir = "logs/image" + os.makedirs(log_dir, exist_ok=True) + + # 保存结果图像 + result_img_path = os.path.join(log_dir, f"center_based_dual_track_{timestamp}.jpg") + cv2.imwrite(result_img_path, result_img) + + # 保存原图 + orig_img_path = os.path.join(log_dir, f"center_based_dual_track_orig_{timestamp}.jpg") + cv2.imwrite(orig_img_path, img) + + info(f"保存中心线基础双轨迹线检测结果图像到: {result_img_path}", "日志") + info(f"保存原始图像到: {orig_img_path}", "日志") + + # 保存文本日志信息 + log_info = { + "timestamp": timestamp, + "center_point": (int(bottom_center_x), height), + "deviation": float(deviation), + "left_track_mid_x": float(left_line[1]), + "right_track_mid_x": float(right_line[1]), + "track_width": float(right_line[1] - left_line[1]), + "center_slope": float(center_slope), + } + info(f"中心线基础双轨迹线检测结果: {log_info}", "日志") + + # 创建左右轨迹线和中心线信息 + left_track_info = { + "line": (left_x1, left_y1, left_x2, left_y2), + "slope": left_line[3], + "x_mid": left_line[1] + } + + right_track_info = { + "line": (right_x1, right_y1, right_x2, right_y2), + "slope": right_line[3], + "x_mid": right_line[1] + } + + center_info = { + "point": (int(bottom_center_x), height), + "deviation": float(deviation), + "slope": float(center_slope), + "is_vertical": abs(center_slope) > 5.0, + "track_width": float(right_line[1] - left_line[1]) + } + + return center_info, left_track_info, right_track_info + +def auto_detect_dual_track_lines(image, observe=False, delay=1000, save_log=True, max_retries=2, use_center_based=True): + """ + 自动检测双轨迹线，使用指定方法 + + 参数: + image: 输入图像，可以是文件路径或者已加载的图像数组 + observe: 是否输出中间状态信息和可视化结果，默认为False + delay: 展示每个步骤的等待时间(毫秒) + save_log: 是否保存日志和图像 + max_retries: 最大重试次数，用于处理复杂情况 + use_center_based: 是否使用基于中心线的检测方法，默认为True + + 返回: + tuple: (中心线信息, 左轨迹线信息, 右轨迹线信息) + """ + # 根据参数选择使用的检测方法 + if use_center_based: + info("使用中心线基础检测方法", "检测") + result = detect_center_based_dual_track_lines(image, observe, delay, save_log) + else: + info("使用传统检测方法", "检测") + result = detect_dual_track_lines(image, observe, delay, save_log, stone_path_mode=False) + + # 检查结果是否成功 + if result[0] is not None: + info("轨迹线检测成功", "检测") + return result + + # 如果失败且还有重试次数，尝试增强图像后重新检测 + if max_retries > 0: + warning(f"检测失败，尝试调整参数重新检测 (剩余重试次数: {max_retries})", "检测") + + # 对图像进行预处理以增强黄线检测 + if isinstance(image, str): + img = cv2.imread(image) + else: + img = image.copy() + + if img is not None: + # 增强图像对比度 + lab = cv2.cvtColor(img, cv2.COLOR_BGR2LAB) + l, a, b = cv2.split(lab) + clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8, 8)) + l = clahe.apply(l) + lab = cv2.merge((l, a, b)) + enhanced_img = cv2.cvtColor(lab, cv2.COLOR_LAB2BGR) + + # 使用增强后的图像重新尝试 + return auto_detect_dual_track_lines(enhanced_img, observe, delay, save_log, max_retries-1, use_center_based) + + error("轨迹线检测失败", "检测") + return None, None, None