diff --git a/src/lane_visualize/main.py b/src/lane_visualize/main.py index c46730aec4..dde08e0182 100644 --- a/src/lane_visualize/main.py +++ b/src/lane_visualize/main.py @@ -3,67 +3,85 @@ import sys from collections import deque +# ========================================== +# 👇 请在这里填入你在 Tuner 中调出的“最好”的数值 +# ========================================== +CANNY_LOW = 50 # 你的 Canny Low +CANNY_HIGH = 150 # 你的 Canny High + +ROI_TOP_WIDTH = 0.40 # 你的 ROI Top W (例如滑动条是40,这里写 0.40) +ROI_HEIGHT_POS = 0.60 # 你的 ROI Height (例如滑动条是60,这里写 0.60) + +HOUGH_THRESH = 20 # 你的 Hough Thresh +MIN_LINE_LEN = 20 # 你的 Min Length +MAX_LINE_GAP = 100 # 你的 Max Gap +# ========================================== + class LaneDetector: def __init__(self): - # --- 1. 放宽霍夫变换参数 --- - # 即使是断断续续的线也能被连起来 - self.rho = 2 # 稍微降低精度以换取更多检测 - self.theta = np.pi / 180 - self.threshold = 15 # 极低阈值:只要有15个点在一条线上就认为是线 - self.min_line_length = 10 # 允许非常短的线段 - self.max_line_gap = 50 # 允许线段之间有缺口 - - # 历史缓存 + # 历史缓存 (用于平滑防抖) self.left_lines_buffer = deque(maxlen=10) self.right_lines_buffer = deque(maxlen=10) self.vertices = None - def process_image_pipeline(self, img): - """ - 处理流水线,返回边缘图以供调试 - """ - # 1. 灰度化 (放弃HLS颜色过滤,改用经典的鲁棒灰度法) - gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) - - # 2. 提高对比度 (直方图均衡化),应对阴天或昏暗环境 - clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8)) - enhanced_gray = clahe.apply(gray) - - # 3. 高斯模糊,去噪 - blur = cv2.GaussianBlur(enhanced_gray, (5, 5), 0) - - # 4. Canny 边缘检测 (动态阈值) - # 自动计算图像的中值亮度,根据亮度调整阈值 - v = np.median(blur) - lower = int(max(0, (1.0 - 0.33) * v)) - upper = int(min(255, (1.0 + 0.33) * v)) - edges = cv2.Canny(blur, lower, upper) - - # 5. ROI 区域掩码 - roi_edges = self.region_of_interest(edges) - - return roi_edges - def region_of_interest(self, img): mask = np.zeros_like(img) if self.vertices is None: height, width = img.shape - # 定义一个更宽的梯形,防止漏掉旁边的线 - self.vertices = np.array([[ - (width * 0.1, height), # 左下 - (width * 0.4, height * 0.6), # 左上 - (width * 0.6, height * 0.6), # 右上 - (width * 0.9, height) # 右下 - ]], dtype=np.int32) + + # 使用填入的参数计算梯形 + top_w = width * ROI_TOP_WIDTH + top_x_center = width * 0.5 + + # 梯形四个顶点 + bl = (0, height) # 左下 + tl = (int(top_x_center - top_w/2), int(height * ROI_HEIGHT_POS)) # 左上 + tr = (int(top_x_center + top_w/2), int(height * ROI_HEIGHT_POS)) # 右上 + br = (width, height) # 右下 + + self.vertices = np.array([[bl, tl, tr, br]], dtype=np.int32) + cv2.fillPoly(mask, self.vertices, 255) return cv2.bitwise_and(img, mask) + def process_frame(self, frame): + if frame is None: return None + + # 1. 图像预处理 + gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY) + blur = cv2.GaussianBlur(gray, (5, 5), 0) + + # 2. 边缘检测 (使用你的参数) + edges = cv2.Canny(blur, CANNY_LOW, CANNY_HIGH) + + # 3. ROI 裁剪 + roi = self.region_of_interest(edges) + + # 4. 霍夫变换 (使用你的参数) + lines = cv2.HoughLinesP(roi, 1, np.pi/180, HOUGH_THRESH, + minLineLength=MIN_LINE_LEN, + maxLineGap=MAX_LINE_GAP) + + # 5. 计算平均线 + left_raw, right_raw = self.average_slope_intercept(lines) + + # 6. 平滑处理 + left_avg, right_avg = self.smooth_lines(left_raw, right_raw) + + # 7. 绘制 + height = frame.shape[0] + y_min = int(height * ROI_HEIGHT_POS) + 50 # 稍微画低一点,不要画到消失点 + y_max = height + + left_pts = self.make_line_points(left_avg, y_min, y_max) + right_pts = self.make_line_points(right_avg, y_min, y_max) + + return self.draw_lane(frame, left_pts, right_pts) + def average_slope_intercept(self, lines): left_lines = [] right_lines = [] - - if lines is None: - return None, None + if lines is None: return None, None for line in lines: x1, y1, x2, y2 = line[0] @@ -72,26 +90,21 @@ def average_slope_intercept(self, lines): slope = (y2 - y1) / (x2 - x1) intercept = y1 - slope * x1 - # 放宽斜率限制:只要不是绝对水平或绝对垂直都算 - # 很多弯道斜率会很小 - if abs(slope) < 0.1 or abs(slope) > 10: + # 斜率过滤:排除水平线和垂直线 + if abs(slope) < 0.3 or abs(slope) > 10: continue - if slope < 0: - left_lines.append((slope, intercept)) - else: - right_lines.append((slope, intercept)) + if slope < 0: left_lines.append((slope, intercept)) + else: right_lines.append((slope, intercept)) left_avg = np.average(left_lines, axis=0) if left_lines else None right_avg = np.average(right_lines, axis=0) if right_lines else None return left_avg, right_avg def smooth_lines(self, left_current, right_current): - if left_current is not None: - self.left_lines_buffer.append(left_current) - if right_current is not None: - self.right_lines_buffer.append(right_current) - + if left_current is not None: self.left_lines_buffer.append(left_current) + if right_current is not None: self.right_lines_buffer.append(right_current) + left_smooth = np.average(self.left_lines_buffer, axis=0) if self.left_lines_buffer else None right_smooth = np.average(self.right_lines_buffer, axis=0) if self.right_lines_buffer else None return left_smooth, right_smooth @@ -104,39 +117,7 @@ def make_line_points(self, line, y_min, y_max): x_min = int((y_min - intercept) / slope) x_max = int((y_max - intercept) / slope) return ((x_min, y_min), (x_max, y_max)) - except: - return None - - def process_frame(self, frame): - if frame is None: return None, None - - # 获取调试用的边缘图 - edges = self.process_image_pipeline(frame) - - # 霍夫检测 - lines = cv2.HoughLinesP(edges, self.rho, self.theta, self.threshold, - np.array([]), minLineLength=self.min_line_length, - maxLineGap=self.max_line_gap) - - left_raw, right_raw = self.average_slope_intercept(lines) - left_avg, right_avg = self.smooth_lines(left_raw, right_raw) - - height = frame.shape[0] - y_min = int(height * 0.65) - y_max = height - - left_pts = self.make_line_points(left_avg, y_min, y_max) - right_pts = self.make_line_points(right_avg, y_min, y_max) - - result_frame = self.draw_lane(frame, left_pts, right_pts) - - # 将边缘图转为彩色以便显示 - debug_frame = cv2.cvtColor(edges, cv2.COLOR_GRAY2BGR) - - # 在调试图上画出ROI区域,方便你看是否对准了 - cv2.polylines(debug_frame, [self.vertices], True, (0, 0, 255), 2) - - return result_frame, debug_frame + except: return None def draw_lane(self, img, left_pts, right_pts): lane_img = np.zeros_like(img) @@ -146,40 +127,30 @@ def draw_lane(self, img, left_pts, right_pts): if left_pts: cv2.line(lane_img, left_pts[0], left_pts[1], (255, 0, 0), 10) if right_pts: cv2.line(lane_img, right_pts[0], right_pts[1], (0, 0, 255), 10) - return cv2.addWeighted(img, 1, lane_img, 0.3, 0) def main(): - detector = LaneDetector() - input_source = sys.argv[1] if len(sys.argv) > 1 else 0 - + # 自动读取 sample.hevc 或摄像头 + input_source = sys.argv[1] if len(sys.argv) > 1 else "sample.hevc" cap = cv2.VideoCapture(input_source) if not cap.isOpened(): - print("无法打开视频") + print("无法打开视频,请检查路径。") return - print("调试模式已启动:请观察 'Debug View' 窗口") + detector = LaneDetector() + print("正在运行... 按 'q' 退出") while True: ret, frame = cap.read() - if not ret: break + if not ret: + # 循环播放 + cap.set(cv2.CAP_PROP_POS_FRAMES, 0) + continue - # 获取结果和调试图 - result, debug_view = detector.process_frame(frame) - - # 拼接图像显示 (左右并排) - # 将原图缩小一点以便并排显示 - h, w = frame.shape[:2] - small_result = cv2.resize(result, (w//2, h//2)) - small_debug = cv2.resize(debug_view, (w//2, h//2)) - - # 纵向拼接:上面是结果,下面是调试(黑白边缘图) - combined = np.vstack((small_result, small_debug)) - - cv2.imshow('Lane Detection Debugger', combined) + result = detector.process_frame(frame) + cv2.imshow('Final Lane Detection', result) - # 慢动作播放 (50ms延时),方便看清 - if cv2.waitKey(50) & 0xFF == ord('q'): + if cv2.waitKey(20) & 0xFF == ord('q'): break cap.release()