diff --git a/src/Autonomous_Navigation/en_class.py b/src/Autonomous_Navigation/en_class.py index bd6db1672..24fb1b9b6 100644 --- a/src/Autonomous_Navigation/en_class.py +++ b/src/Autonomous_Navigation/en_class.py @@ -1,148 +1,313 @@ -# src/environment_classifier.py +#!/usr/bin/env python +# -*- coding: utf-8 -*- + +""" +环境分类器模块 - 用于无人机视觉导航系统 +基于手工特征（LAB颜色、纹理、颜色比例、空间网格）和规则判断， +无需机器学习库，仅依赖 OpenCV 和 NumPy。 +""" + import cv2 import numpy as np import random +from typing import Dict, Tuple, List, Optional class EnvironmentClassifier: """增强版环境分类器（仅依赖OpenCV和numpy）""" - def __init__(self): - self.environments = [ - "Ruins", "Building", "Forest", "Road", - "Sky", "Water", "Fire", "Animal", "Vehicle" - ] - self.weights = { - "Ruins": 0.35, "Building": 0.20, "Forest": 0.15, - "Road": 0.10, "Sky": 0.08, "Water": 0.05, - "Fire": 0.02, "Animal": 0.03, "Vehicle": 0.02 - } - self.thresholds = { - 'fire_red': 0.25, 'fire_bright': 200, 'fire_grad': 15, - 'sky_blue': 0.3, 'sky_l_mean': 180, - 'water_blue': 0.25, 'water_edges': 0.03, - 'forest_green': 0.3, 'forest_grad': 20, - 'ruins_edges': 0.07, 'ruins_variance': 1200, 'ruins_bright': 120, - 'building_edges': 0.05, 'building_bright': 100, 'building_green': 0.1, - 'road_edges': 0.02, 'road_bright_low': 100, 'road_bright_high': 200, 'road_color': 0.1, - 'animal_red': 0.15, 'animal_edges': 0.04 - } - - def classify(self, image): + # 环境类别列表（按优先级或常用顺序） + ENVIRONMENTS: List[str] = [ + "Ruins", "Building", "Forest", "Road", + "Sky", "Water", "Fire", "Animal", "Vehicle" + ] + + # 基础权重（用于加权随机回退） + BASE_WEIGHTS: Dict[str, float] = { + "Ruins": 0.35, "Building": 0.20, "Forest": 0.15, + "Road": 0.10, "Sky": 0.08, "Water": 0.05, + "Fire": 0.02, "Animal": 0.03, "Vehicle": 0.02 + } + + # 规则判断阈值（集中管理，便于调优） + THRESHOLDS: Dict[str, float] = { + # 火灾 + 'fire_red': 0.25, + 'fire_bright': 200, + 'fire_grad': 15, + # 天空 + 'sky_blue': 0.3, + 'sky_l_mean': 180, + # 水域 + 'water_blue': 0.25, + 'water_edges': 0.03, + # 森林 + 'forest_green': 0.3, + 'forest_grad': 20, + # 废墟 + 'ruins_edges': 0.07, + 'ruins_variance': 1200, + 'ruins_bright': 120, + # 建筑 + 'building_edges': 0.05, + 'building_bright': 100, + 'building_green': 0.1, + # 道路 + 'road_edges': 0.02, + 'road_bright_low': 100, + 'road_bright_high': 200, + 'road_color': 0.1, + # 动物/车辆（简化） + 'animal_red': 0.15, + 'animal_edges': 0.04 + } + + def __init__(self) -> None: + """初始化分类器：设置环境列表、权重、阈值""" + self.environments = self.ENVIRONMENTS.copy() + self.weights = self.BASE_WEIGHTS.copy() + self.thresholds = self.THRESHOLDS.copy() + + # ------------------------------------------------------------------ + # 公共接口 + # ------------------------------------------------------------------ + def classify(self, image: Optional[np.ndarray]) -> Tuple[str, float]: + """ + 对输入图像进行环境分类 + + :param image: BGR 格式的图像（OpenCV 读取） + :return: (环境名称, 置信度) + """ if image is None: return "Unknown", 0.0 + + # 提取特征 features = self._extract_features(image) - env, conf = self._rule_based(features) + + # 优先使用规则判断 + env, conf = self._rule_based_classify(features) + + # 若规则无法判断，则使用加权随机回退 if env == "Unknown": env, conf = self._weighted_random(features) + return env, conf - def _extract_features(self, image): + # ------------------------------------------------------------------ + # 特征提取（私有方法） + # ------------------------------------------------------------------ + def _extract_features(self, image: np.ndarray) -> Dict[str, float]: + """ + 从图像中提取丰富的特征向量 + + 包括： + - LAB 颜色通道的均值和标准差 (6维) + - 纹理特征：Sobel梯度均值/标准差、Laplacian方差、Canny边缘密度 + - 全局颜色比例：HSV空间下的蓝、绿、红色像素比例 + - 天空检测标志 + - 3×3 网格颜色比例 (27维) + + :param image: BGR 图像 + :return: 特征字典 + """ h, w = image.shape[:2] + + # 转换颜色空间（一次性转换，避免重复计算） gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV) lab = cv2.cvtColor(image, cv2.COLOR_BGR2LAB) - f = {} - # LAB + + features = {} + + # ----- 1. LAB 颜色统计 ----- l_mean, l_std = cv2.meanStdDev(lab[:, :, 0]) a_mean, a_std = cv2.meanStdDev(lab[:, :, 1]) b_mean, b_std = cv2.meanStdDev(lab[:, :, 2]) - f['l_mean'] = l_mean[0][0] - f['l_std'] = l_std[0][0] - f['a_mean'] = a_mean[0][0] - f['a_std'] = a_std[0][0] - f['b_mean'] = b_mean[0][0] - f['b_std'] = b_std[0][0] - # 纹理 + features['l_mean'] = l_mean[0][0] + features['l_std'] = l_std[0][0] + features['a_mean'] = a_mean[0][0] + features['a_std'] = a_std[0][0] + features['b_mean'] = b_mean[0][0] + features['b_std'] = b_std[0][0] + + # ----- 2. 纹理特征 ----- + # Sobel 梯度幅值 sobelx = cv2.Sobel(gray, cv2.CV_64F, 1, 0, ksize=3) sobely = cv2.Sobel(gray, cv2.CV_64F, 0, 1, ksize=3) mag = np.sqrt(sobelx ** 2 + sobely ** 2) - f['grad_mean'] = np.mean(mag) - f['grad_std'] = np.std(mag) + features['grad_mean'] = np.mean(mag) + features['grad_std'] = np.std(mag) + + # Laplacian 方差（反映图像清晰度/纹理复杂度） laplacian = cv2.Laplacian(gray, cv2.CV_64F) - f['lap_var'] = np.var(laplacian) + features['lap_var'] = np.var(laplacian) + + # Canny 边缘密度 edges = cv2.Canny(gray, 50, 150) - f['edge_density'] = np.sum(edges > 0) / (h * w) - # 颜色比例 + features['edge_density'] = np.sum(edges > 0) / (h * w) + + # ----- 3. 全局颜色比例（HSV 空间）----- + # 蓝色范围 blue_mask = cv2.inRange(hsv, (100, 50, 50), (130, 255, 255)) - f['blue_ratio'] = np.sum(blue_mask > 0) / (h * w) + features['blue_ratio'] = np.sum(blue_mask > 0) / (h * w) + + # 绿色范围 green_mask = cv2.inRange(hsv, (40, 50, 50), (80, 255, 255)) - f['green_ratio'] = np.sum(green_mask > 0) / (h * w) + features['green_ratio'] = np.sum(green_mask > 0) / (h * w) + + # 红色范围（HSV中红色分布在0~10和170~180两个区间） red_mask1 = cv2.inRange(hsv, (0, 50, 50), (10, 255, 255)) red_mask2 = cv2.inRange(hsv, (170, 50, 50), (180, 255, 255)) red_mask = cv2.bitwise_or(red_mask1, red_mask2) - f['red_ratio'] = np.sum(red_mask > 0) / (h * w) - f['brightness'] = np.mean(gray) - f['gray_variance'] = np.var(gray) - # 天空检测 + features['red_ratio'] = np.sum(red_mask > 0) / (h * w) + + # 全局亮度与灰度方差 + features['brightness'] = np.mean(gray) + features['gray_variance'] = np.var(gray) + + # ----- 4. 天空检测（基于顶部蓝色比例和LAB亮度）----- if h > 10: top_blue = np.sum(blue_mask[:h // 3, :] > 0) / (w * h // 3) bottom_blue = np.sum(blue_mask[2 * h // 3:, :] > 0) / (w * h // 3) - f['is_sky'] = (top_blue > 0.25) and (top_blue > bottom_blue * 1.8) and (f['l_mean'] > 180) + features['is_sky'] = ( + top_blue > 0.25 and + top_blue > bottom_blue * 1.8 and + features['l_mean'] > 180 + ) else: - f['is_sky'] = False - # 3x3网格（可选，增加特征维度） + features['is_sky'] = False + + # ----- 5. 3×3 网格颜色比例（空间分布信息）----- grid_h, grid_w = h // 3, w // 3 for i in range(3): for j in range(3): + # 提取当前网格的 ROI roi = image[i * grid_h:(i + 1) * grid_h, j * grid_w:(j + 1) * grid_w] - if roi.size == 0: continue + if roi.size == 0: + continue + roi_hsv = cv2.cvtColor(roi, cv2.COLOR_BGR2HSV) + + # 网格内蓝色比例 blue_roi = cv2.inRange(roi_hsv, (100, 50, 50), (130, 255, 255)) - f[f'blue_grid_{i}_{j}'] = np.sum(blue_roi > 0) / (grid_h * grid_w) + features[f'blue_grid_{i}_{j}'] = np.sum(blue_roi > 0) / (grid_h * grid_w) + + # 网格内绿色比例 green_roi = cv2.inRange(roi_hsv, (40, 50, 50), (80, 255, 255)) - f[f'green_grid_{i}_{j}'] = np.sum(green_roi > 0) / (grid_h * grid_w) + features[f'green_grid_{i}_{j}'] = np.sum(green_roi > 0) / (grid_h * grid_w) + + # 网格内红色比例 red_roi1 = cv2.inRange(roi_hsv, (0, 50, 50), (10, 255, 255)) red_roi2 = cv2.inRange(roi_hsv, (170, 50, 50), (180, 255, 255)) red_roi = cv2.bitwise_or(red_roi1, red_roi2) - f[f'red_grid_{i}_{j}'] = np.sum(red_roi > 0) / (grid_h * grid_w) - return f - - def _rule_based(self, f): - t = self.thresholds - if (f.get('red_ratio', 0) > t['fire_red'] and f.get('brightness', 0) > t['fire_bright'] and f.get('grad_mean', - 0) < t[ - 'fire_grad']): + features[f'red_grid_{i}_{j}'] = np.sum(red_roi > 0) / (grid_h * grid_w) + + return features + + # ------------------------------------------------------------------ + # 规则分类（私有方法） + # ------------------------------------------------------------------ + def _rule_based_classify(self, f: Dict[str, float]) -> Tuple[str, float]: + """ + 基于预设规则进行环境分类（按优先级依次判断） + + :param f: 特征字典 + :return: (环境名称, 置信度) 或 ("Unknown", 0.0) + """ + t = self.thresholds # 本地别名，加速访问 + + # 火灾：红色比例高 + 亮度高 + 纹理模糊（梯度均值低） + if (f.get('red_ratio', 0) > t['fire_red'] and + f.get('brightness', 0) > t['fire_bright'] and + f.get('grad_mean', 0) < t['fire_grad']): return "Fire", 0.75 - if (f.get('blue_ratio', 0) > t['sky_blue'] and f.get('is_sky', False) and f.get('l_mean', 0) > t['sky_l_mean']): + + # 天空：蓝色比例高 + 天空标志 + LAB L 均值高 + if (f.get('blue_ratio', 0) > t['sky_blue'] and + f.get('is_sky', False) and + f.get('l_mean', 0) > t['sky_l_mean']): return "Sky", 0.85 - if (f.get('blue_ratio', 0) > t['water_blue'] and not f.get('is_sky', False) and f.get('edge_density', 0) < t[ - 'water_edges']): + + # 水域：蓝色比例较高 + 非天空 + 边缘密度低 + if (f.get('blue_ratio', 0) > t['water_blue'] and + not f.get('is_sky', False) and + f.get('edge_density', 0) < t['water_edges']): return "Water", 0.70 - if (f.get('green_ratio', 0) > t['forest_green'] and f.get('grad_mean', 0) > t['forest_grad']): + + # 森林：绿色比例高 + 梯度均值高（纹理丰富） + if (f.get('green_ratio', 0) > t['forest_green'] and + f.get('grad_mean', 0) > t['forest_grad']): return "Forest", 0.80 - if (f.get('edge_density', 0) > t['ruins_edges'] and f.get('gray_variance', 0) > t['ruins_variance'] and f.get( - 'brightness', 0) < t['ruins_bright']): + + # 废墟：边缘密度高 + 灰度方差高 + 亮度低 + if (f.get('edge_density', 0) > t['ruins_edges'] and + f.get('gray_variance', 0) > t['ruins_variance'] and + f.get('brightness', 0) < t['ruins_bright']): return "Ruins", 0.85 - if (f.get('edge_density', 0) > t['building_edges'] and f.get('brightness', 0) < t['building_bright'] and f.get( - 'green_ratio', 0) < t['building_green']): + + # 建筑：边缘密度较高 + 亮度低 + 绿色比例低 + if (f.get('edge_density', 0) > t['building_edges'] and + f.get('brightness', 0) < t['building_bright'] and + f.get('green_ratio', 0) < t['building_green']): return "Building", 0.75 - if (f.get('edge_density', 0) < t['road_edges'] and t['road_bright_low'] < f.get('brightness', 0) < t[ - 'road_bright_high'] and - f.get('blue_ratio', 0) < t['road_color'] and f.get('green_ratio', 0) < t['road_color'] and f.get( - 'red_ratio', 0) < t['road_color']): + + # 道路：边缘密度低 + 中等亮度 + 各颜色比例低 + if (f.get('edge_density', 0) < t['road_edges'] and + t['road_bright_low'] < f.get('brightness', 0) < t['road_bright_high'] and + f.get('blue_ratio', 0) < t['road_color'] and + f.get('green_ratio', 0) < t['road_color'] and + f.get('red_ratio', 0) < t['road_color']): return "Road", 0.70 - if (f.get('red_ratio', 0) > t['animal_red'] and f.get('edge_density', 0) > t['animal_edges']): + + # 动物/车辆（简化，特征相似）：红色比例较高 + 边缘密度较高 + if (f.get('red_ratio', 0) > t['animal_red'] and + f.get('edge_density', 0) > t['animal_edges']): return "Vehicle", 0.60 + + # 无法确定 return "Unknown", 0.0 - def _weighted_random(self, f): - adj = self.weights.copy() + # ------------------------------------------------------------------ + # 加权随机回退（私有方法） + # ------------------------------------------------------------------ + def _weighted_random(self, f: Dict[str, float]) -> Tuple[str, float]: + """ + 当规则无法确定时，根据图像特征调整基础权重，然后随机选择一个环境 + + :param f: 特征字典 + :return: (环境名称, 置信度) + """ + # 复制基础权重 + adj_weights = self.weights.copy() + + # 获取关键特征 blue = f.get('blue_ratio', 0) green = f.get('green_ratio', 0) edges = f.get('edge_density', 0) + + # 根据特征调整权重 if blue > 0.2: - adj["Sky"] *= 1.5 + adj_weights["Sky"] *= 1.5 if blue > 0.3: - adj["Water"] *= 0.5 + adj_weights["Water"] *= 0.5 # 减少水权重，避免误判 + if green > 0.15: - adj["Forest"] *= 2.0 + adj_weights["Forest"] *= 2.0 + if edges > 0.05: - adj["Ruins"] *= 1.8 - total = sum(adj.values()) - probs = [adj[env] / total for env in self.environments] if total > 0 else [1 / len(self.environments)] * len( - self.environments) + adj_weights["Ruins"] *= 1.8 + + # 归一化，计算概率分布 + total = sum(adj_weights.values()) + if total > 0: + probs = [adj_weights[env] / total for env in self.environments] + else: + probs = [1.0 / len(self.environments)] * len(self.environments) + + # 随机选择环境 env = random.choices(self.environments, weights=probs)[0] + + # 计算基础置信度，并根据特征微调 conf = 0.6 if env == "Ruins" and edges > 0.04: conf += 0.15 @@ -150,4 +315,5 @@ def _weighted_random(self, f): conf += 0.1 elif env == "Sky" and blue > 0.2: conf += 0.1 - return env, min(conf, 0.9) \ No newline at end of file + + return env, min(conf, 0.9) # 置信度不超过0.9 \ No newline at end of file diff --git a/src/Autonomous_Navigation/scripts/123.bat b/src/Autonomous_Navigation/scripts/123.bat new file mode 100644 index 000000000..d7b4bb860 --- /dev/null +++ b/src/Autonomous_Navigation/scripts/123.bat @@ -0,0 +1,52 @@ +@echo off +cls + +echo ============================================ +echo DRONE VISION NAVIGATION SYSTEM +echo ============================================ +echo. + +echo Step 1: Checking simulator... +tasklist | findstr "AbandonedPark.exe" > nul +if errorlevel 1 ( + echo Simulator not running! + echo Please run AbandonedPark.exe manually + echo Waiting 20 seconds... + timeout /t 20 /nobreak > nul +) else ( + echo ✓ Simulator is running +) + +echo. +echo Step 2: Activating Python environment... +if exist "venv\Scripts\activate.bat" ( + call venv\Scripts\activate.bat + echo ✓ Virtual environment activated +) else ( + echo Creating virtual environment... + python -m venv venv + call venv\Scripts\activate.bat + echo Installing dependencies... + pip install airsim opencv-python numpy +) + +echo. +echo Step 3: Checking Python packages... +python -c "import airsim; print('✓ AirSim installed')" 2>nul +if errorlevel 1 ( + echo Installing AirSim... + pip install airsim +) + +python -c "import cv2; print('✓ OpenCV installed')" 2>nul +if errorlevel 1 ( + echo Installing OpenCV... + pip install opencv-python +) + +echo. +echo Step 4: Starting drone system... +python drone_vision_system_en.py + +echo. +pause \ No newline at end of file