基于Python对道路上的车辆进行检测和识别源代码+文档说明+演示视频(高分毕业设计)

## 车辆检测 将会进行如下操作： 1. 定义图像特征提取方 2. 读入车辆和非车辆图像数据 3. 数据预处理，shuffle，标准化 4. 训练模型，使用的是svm 5. 测试模型 6. 使用滑动窗口和多尺度的图像大小在图像上检测车辆 - 滑动窗口 - 多尺度图像 - heatmap - label - box center 7. 检测视频中行驶的车辆 ​ **总结：车辆检测主要是两方面的内容，1:物体识别 2:滑动窗口** - 汽车和非汽车图片展示 ```python import glob import matplotlib.pyplot as plt vehicle_lable = 1 non_vehicle_label = 0 # 读入数据 vehicles_file_paths = glob.glob("./vehicles/*/*.png") non_vehicles_file_paths = glob.glob("./non-vehicles/*/*.png") print("汽车样本数量：" , len(vehicles_file_paths)) print("非汽车样本数量：" , len(non_vehicles_file_paths)) # 显示汽车和非汽车图片 vehicle_img = mpimg.imread(vehicles_file_paths[0]) non_vehicle_img = mpimg.imread(non_vehicles_file_paths[0]) fig, (ax1, ax2) = plt.subplots(1, 2 ,figsize=(8, 20)) ax1.imshow(vehicle_img) ax1.set_title("vehicle") ax2.imshow(non_vehicle_img) ax2.set_title("non_vehicle") plt.show() ```  - 提取HOG特征 > hog参数说明： > > orientations：每个cell直方图的方向的个数 > > pix_per_cell：每个cell像素个数 > > cells_per_block：每个block的cell个数 ```python import numpy as np import cv2 import matplotlib.image as mpimg def convert_color(img, conv): """ 颜色空间转换 """ return cv2.cvtColor(img, conv) def get_hog_features(img, orient, pix_per_cell, cell_per_block, vis=False, feature_vec=True): """ 提取图像的hog特征 """ if vis == True: features, hog_image = hog(img, orientations=orient, pixels_per_cell=(pix_per_cell, pix_per_cell), cells_per_block=(cell_per_block, cell_per_block), block_norm= 'L2-Hys', transform_sqrt=False, visualise=vis, feature_vector=feature_vec) return features, hog_image else: features = hog(img, orientations=orient, pixels_per_cell=(pix_per_cell, pix_per_cell), cells_per_block=(cell_per_block, cell_per_block), block_norm= 'L2-Hys', transform_sqrt=False, visualise=vis, feature_vector=feature_vec) return features # 测试提取HOG特征 vehicle_img = mpimg.imread(vehicles_file_paths[0]) non_vehicle_img = mpimg.imread(non_vehicles_file_paths[0]) v_hog_feats,v_hog_imgs = get_hog_features(vehicle_img[:, :, 0], 9, 8, 2, vis=True, feature_vec=False) n_hog_feats,n_hog_imgs = get_hog_features(non_vehicle_img[:, :, 0], 9, 8, 2, vis=True, feature_vec=False) # 显示汽车和非汽车图片 vehicle_img = mpimg.imread(vehicles_file_paths[0]) non_vehicle_img = mpimg.imread(non_vehicles_file_paths[0]) fig, (ax1, ax2) = plt.subplots(1, 2 ,figsize=(8, 20)) ax1.imshow(v_hog_imgs, cmap="gray") ax1.set_title("vehicle") ax2.imshow(n_hog_imgs, cmap="gray") ax2.set_title("non_vehicle") plt.show() ```  - 提取空间特征和颜色直方图特征，且把之上的特征进行合并。 ```python def bin_spatial(img, size=(32, 32)): """ 提取图像的空间特征 """ color1 = cv2.resize(img[:,:,0], size).ravel() color2 = cv2.resize(img[:,:,1], size).ravel() color3 = cv2.resize(img[:,:,2], size).ravel() return np.hstack((color1, color2, color3)) def color_hist(img, nbins=32): """ 颜色直方图 """ channel1_hist = np.histogram(img[:,:,0], bins=nbins) channel2_hist = np.histogram(img[:,:,1], bins=nbins) channel3_hist = np.histogram(img[:,:,2], bins=nbins) hist_features = np.concatenate((channel1_hist[0], channel2_hist[0], channel3_hist[0])) return hist_features def extract_img_features(img): """ 把之上的图像处理方法进行综合，且对结果进行归一化处理 """ img = cv2.cvtColor(img, cv2.COLOR_RGB2YCrCb) hist_feature = color_hist(img) spatial_feature = bin_spatial(img) ch1 = img[:, :, 0] ch2 = img[:, :, 1] ch3 = img[:, :, 2] ch1_hog_feature = get_hog_features(ch1, 9, 8, 2) ch2_hog_feature = get_hog_features(ch2, 9, 8, 2) ch3_hog_feature = get_hog_features(ch3, 9, 8, 2) hog_features = np.hstack((ch1_hog_feature, ch2_hog_feature, ch3_hog_feature)) features = np.concatenate((hist_feature, spatial_feature, hog_features)) return features ``` - 根据上边的图像特征提取方法，对所有的数据进行特征提取 ```python X = [] y = [] for path in vehicles_file_paths: img = mpimg.imread(path) features = extract_img_features(img) X.append(features) y.append(vehicle_lable) for path in non_vehicles_file_paths: img = mpimg.imread(path) features = extract_img_features(img) X.append(features) y.append(non_vehicle_label) ``` - 数据预处理 > shuffle —> train_test_split—>StandardScaler > > 注意：数据标准化的时候，使用训练数据作为fit的对象。 ```python from skimage.feature import hog from sklearn.model_selection import train_test_split from sklearn.utils import shuffle from sklearn.preprocessing import StandardScaler # shuffle X, y = shuffle(X, y) # split rand_state = np.random.randint(0, 100) X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=rand_state) # scaller X_scaler = StandardScaler().fit(X_train) X_train = X_scaler.transform(X_train) X_test = X_scaler.transform(X_test) ``` - 训练模型 ```python from sklearn.svm import LinearSVC svm = LinearSVC() svm.fit(X_train, y_train) ``` - 测试模型 > 最后模型的准确率为：0.986204954955 ```python from sklearn.metrics import accuracy_score predicts = svm.predict(X_test) score = accuracy_score(predicts, y_test) print("在测试数据上的准确率为：" + score) ``` - 从图片中检测汽车 > **滑动窗口技术是关键** > > 因为不知道检测物体的大小，远的地方车辆看起来小，近的地方看起来大。为了得到合适的检测数据，需要图片进行适当的缩放。但是这样带来的后果是，计算量会急剧增加。可以采查找车辆只在图像的下半部分，因为上半部分是天空。为了提高效率，先对整幅图像进行HOG特征提取，然后再根据需要进行特征分割。 ```python def find_cars(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins): """ 检测当前图像中的所有可能的汽车位置 """ boxs = [] draw_img = np.copy(img) img = img.astype(np.float32)/255 img_tosearch = img[ystart:ystop,:,:] ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2YCrCb) if scale != 1: imshape = ctrans_tosearch.shape ctrans_tosearch = cv2.resize(ctrans_tosearch, (np.int(imshape[1]/scale), np.int(imshape[0]/scale))) ch1 = ctrans_tosearch[:,:,0] ch2 = ctrans_tosearch[:,:,1] ch3 = ctrans_tosearch[:,:,2] # Define blocks and steps as above nxblocks = (ch1.shape[1] // pix_per_cell) - cell_per_block + 1 nyblocks = (ch1.shape[0] // pix_per_cell) - cell_per_block + 1 nfeat_per_block = orient*cell_per_block**2 # 64 was the orginal sampling rate, with 8 cells and 8 pix per cell window = 64 nblocks_per_window = (window // pix_per_cell) - cell_pe