import cv2 import numpy as np import os import datetime from utils.base_line_handler import _merge_collinear_lines_iterative 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, max_slope_threshold=2, min_slope_threshold=0.5, min_line_length=0.05, max_line_gap=40): """ 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 max_slope_threshold: 最大斜率阈值 min_slope_threshold: 最小斜率阈值 min_line_length: 最小线段长度 max_line_gap: 最大线段间距 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) # 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) 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") mask_display = cv2.cvtColor(mask, cv2.COLOR_GRAY2BGR) cv2.putText(mask_display, "Step 1: Yellow Mask", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 2) cv2.putText(mask_display, f"Lower: {lower_yellow}", (10, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 255, 0), 1) cv2.putText(mask_display, f"Upper: {upper_yellow}", (10, 55), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 255, 0), 1) cv2.imshow("黄色掩码", mask_display) 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") bottom_roi_display = cv2.cvtColor(bottom_roi, cv2.COLOR_GRAY2BGR) cv2.putText(bottom_roi_display, "Step 1.5: Bottom ROI", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 2) cv2.putText(bottom_roi_display, f"ROI Height: {bottom_roi_height}px ({bottom_roi_height/height*100:.0f}%)", (10, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 255, 0), 1) cv2.imshow("底部区域掩码", bottom_roi_display) cv2.waitKey(delay) # INFO 边缘检测 edges = cv2.Canny(mask, 50, 150, apertureSize=3) if observe: debug("步骤2: 边缘检测", "处理") edges_display = cv2.cvtColor(edges, cv2.COLOR_GRAY2BGR) cv2.putText(edges_display, "Step 2: Edge Detection (Canny)", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 2) cv2.putText(edges_display, "Thresholds: (50, 150)", (10, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 255, 0), 1) cv2.imshow("边缘检测", edges_display) cv2.waitKey(delay) # INFO 霍夫变换检测直线 lines = cv2.HoughLinesP(edges, 1, np.pi/180, threshold=25, minLineLength=width*min_line_length, maxLineGap=max_line_gap) 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() cv2.putText(lines_img, "Step 3: Hough Lines", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2) cv2.putText(lines_img, f"Th:25, MinLen:{width*0.05:.1f}, MaxGap:{max_line_gap}", (10, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 0, 255), 1) 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)，但允许更多倾斜度 if max_slope_threshold > 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)) # vertical_lines = _merge_collinear_lines_iterative(vertical_lines, # min_initial_len=5.0, # max_angle_diff_deg=10.0, # max_ep_gap_abs=max_line_gap / 2.0, # max_ep_gap_factor=0.25, # max_p_dist_abs=max_line_gap / 4.0, # max_p_dist_factor=0.1) if len(vertical_lines) < 2: 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() cv2.putText(v_lines_img, "Step 4: Vertical Lines Filtered", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2) cv2.putText(v_lines_img, f"Min Slope Abs: {min_slope_threshold:.2f}", (10, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 0, 255), 1) 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 # 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") # 创建左右线可视化图像 left_right_img = img.copy() cv2.putText(left_right_img, "Left and Right Candidate Lines", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2) # 绘制左侧候选线（蓝色） for line_info in left_lines: line = line_info[0] x1, y1, x2, y2 = line cv2.line(left_right_img, (x1, y1), (x2, y2), (255, 0, 0), 2) # 显示斜率 mid_x = (x1 + x2) // 2 mid_y = (y1 + y2) // 2 cv2.putText(left_right_img, f"L:{line_info[3]:.2f}", (mid_x, mid_y), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1) # 绘制右侧候选线（红色） for line_info in right_lines: line = line_info[0] x1, y1, x2, y2 = line cv2.line(left_right_img, (x1, y1), (x2, y2), (0, 0, 255), 2) # 显示斜率 mid_x = (x1 + x2) // 2 mid_y = (y1 + y2) // 2 cv2.putText(left_right_img, f"R:{line_info[3]:.2f}", (mid_x, mid_y), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 1) cv2.imshow("左右候选线", left_right_img) cv2.waitKey(delay) # 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 # INFO 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 center_x = width / 2 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 # 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 * 0.05 + # 线段长度 x_score * 0.15 + # 中点位置 center_proximity_score * 0.3 + # 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 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 # 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 # 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) cv2.putText(result_img, "Final Result", (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2) if 'best_pair_score' in locals() and best_pair_score != -1: cv2.putText(result_img, f"Pair Score: {best_pair_score:.2f}", (10, 85), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1) current_y_offset = 105 else: current_y_offset = 85 cv2.putText(result_img, f"L-Slope: {left_line[3]:.2f}, R-Slope: {right_line[3]:.2f}", (10, current_y_offset), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1) current_y_offset += 20 cv2.putText(result_img, f"Track Width: {right_line[1] - left_line[1]:.1f}px", (10, current_y_offset), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1) 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), } 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 } 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") mask_display = cv2.cvtColor(combined_mask, cv2.COLOR_GRAY2BGR) cv2.putText(mask_display, "Step 1: Yellow Mask (Center Based)", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 2) cv2.putText(mask_display, f"Lower: {lower_yellow} + Darker Var.", (10, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 255, 0), 1) cv2.putText(mask_display, f"Upper: {upper_yellow} + Darker Var.", (10, 55), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 255, 0), 1) cv2.imshow("黄色掩码", mask_display) 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") bottom_roi_display = cv2.cvtColor(bottom_roi, cv2.COLOR_GRAY2BGR) cv2.putText(bottom_roi_display, "Step 1.5: Bottom ROI (Center Based)", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 2) cv2.putText(bottom_roi_display, f"ROI Height: {bottom_roi_height}px ({bottom_roi_height/height*100:.0f}%)", (10, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 255, 0), 1) cv2.imshow("底部区域掩码", bottom_roi_display) 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") edges_display = cv2.cvtColor(edges, cv2.COLOR_GRAY2BGR) cv2.putText(edges_display, "Step 2: Edge Detection (Canny - Center Based)", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 2) cv2.putText(edges_display, "Thresholds: (50, 150)", (10, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 255, 0), 1) cv2.imshow("边缘检测", edges_display) 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 (或合并后无直线 - Center Based)", "失败") return None, None, None # 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)，但允许更多倾斜度 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_display = img.copy() cv2.putText(v_lines_img_display, "Step 4: Vertical Lines Filtered (Center Based)", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2) cv2.putText(v_lines_img_display, "Min Slope Abs: 0.50", (10, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 0, 255), 1) # Slope threshold is 0.5 here for line_info in vertical_lines: line, _, _, slope, _ = line_info x1, y1, x2, y2 = line cv2.line(v_lines_img_display, (x1, y1), (x2, y2), (0, 255, 255), 2) # 显示斜率 cv2.putText(v_lines_img_display, 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_display) cv2.waitKey(delay) debug(f"步骤2.5 (Center Based): 初始检测到 {len(vertical_lines)} 条垂直线，尝试合并", "处理") # 合并参数设置 (可以根据需要调整，这里使用与 detect_dual_track_lines 类似的设置) # 注意：这里的 max_line_gap 是 HoughLinesP 本身的参数，不是函数参数 # 如果需要独立于 detect_dual_track_lines 中的 max_line_gap，需要单独定义或传递 current_max_line_gap = 40 # 来自 HoughLinesP min_len_for_merge_cb = 5.0 angle_thresh_deg_cb = 10.0 ep_gap_abs_cb = current_max_line_gap / 2.0 ep_gap_factor_cb = 0.25 p_dist_abs_cb = current_max_line_gap / 4.0 p_dist_factor_cb = 0.1 # vertical_lines = _merge_collinear_lines_iterative(vertical_lines, # min_initial_len=min_len_for_merge_cb, # max_angle_diff_deg=angle_thresh_deg_cb, # max_ep_gap_abs=ep_gap_abs_cb, # max_ep_gap_factor=ep_gap_factor_cb, # max_p_dist_abs=p_dist_abs_cb, # max_p_dist_factor=p_dist_factor_cb) # INFO 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) if observe: debug(f"步骤3.5: 找到 {len(center_candidates)} 条可能的中心线", "处理") center_candidates_img_display = img.copy() cv2.putText(center_candidates_img_display, "Step 3: Center Candidates (Center Based)", (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2) for line_info in center_candidates: line, _, _, slope, _ = line_info x1, y1, x2, y2 = line cv2.line(center_candidates_img_display, (x1, y1), (x2, y2), (0, 255, 255), 2) # 显示斜率 cv2.putText(center_candidates_img_display, f"{slope:.2f}", ((x1+x2)//2, (y1+y2)//2), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1) cv2.imshow("可能的中心线", center_candidates_img_display) cv2.waitKey(delay) # 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 # INFO 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) cv2.putText(result_img, "Final Result (Center Based)", (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2) if 'best_score' in locals() and best_score != -1: cv2.putText(result_img, f"Pair Score: {best_score:.2f}", (10, 85), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1) current_y_offset = 105 else: current_y_offset = 85 cv2.putText(result_img, f"Center Slope: {center_slope:.2f}", (10, current_y_offset), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1) current_y_offset += 20 cv2.putText(result_img, f"L-Slope: {left_line[3]:.2f}, R-Slope: {right_line[3]:.2f}", (10, current_y_offset), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1) current_y_offset += 20 cv2.putText(result_img, f"Track Width: {right_line[1] - left_line[1]:.1f}px", (10, current_y_offset), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1) 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) # 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