简介

p5从车道检测升级为车辆检测，课程目的是检测行驶过程中遇到的车辆，属于目标检测范畴。这篇文章主要介绍如何使用传统计算机视觉方法进行车辆识别，比较基础。当然实现车辆识别的方案还有其他高大的方案，如yolo等，后面我会开另一篇文章介绍基于深度学习的车辆检测方案。

处理流程

本节课程处理流程如下图所示

HOG特征提取

梯度直方图(Histogram of Oriented Gradient)，简称HOG，它是一种特征描述符。其原理是把图像分成很多个部分，然后计算每个部分的梯度，HOG特征描述符能为特征匹配和目标检测提供非常重要的信息。

为了判断是否是车辆，我们需要分别提取车辆的HOG和非车辆物体的HOG。这里我们使用skimage的hog方法进行图像的特征提取，经过特征提取后的效果如下

贴上这部分代码

import cv2
import numpy as np
from skimage.feature import hog
import glob
import matplotlib.pyplot as plt

#读取car data
def load_car_data():
	vehicles = glob.glob('./vehicles/*/*.png')
	return vehicles
#读取non car data
def load_non_car_data():
	non_vehicles = glob.glob('./non-vehicles/*/*.png')
	return non_vehicles

#提取HOG特征
def extract_feature(img,orient,pix_per_cell,cell_per_block,vis=False,feature_vec = True):
	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),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),transform_sqrt=False,visualise=vis,feature_vector=feature_vec)
		return feature
#显示对比
def show_hog():
	vehicles = load_car_data()
	non_vehicles = load_non_car_data()
	vehicle_img = plt.imread(vehicles[5])
	non_vehicle_img = plt.imread(non_vehicles[5])
	_,vehicle_hog_image = extract_feature(vehicle_img[:,:,2],9,8,8,vis=True,feature_vec=True)
	_,non_vehicle_hog_image = extract_feature(non_vehicle_img[:,:,2],9,8,8,vis=True,feature_vec=True)
	f,((ax1,ax2),(ax3,ax4)) = plt.subplots(2,2,figsize=(7,7))
	ax1.imshow(vehicle_img)
	ax1.set_title('car')
	ax2.imshow(vehicle_hog_image)
	ax2.set_title('car hog feature')
	ax3.imshow(non_vehicle_img)
	ax3.set_title('non car')
	ax4.imshow(non_vehicle_hog_image)
	ax4.set_title('non car hog feature')
	plt.show()

svm模型训练

我们刚刚获取了车辆和非车辆的hog特征，接下来用这些特征去训练svm模型。

def split_data():
	car_data = load_car_data()
	car_features = extract_features(car_data)
	non_car_data = load_non_car_data()
	non_car_features = extract_features(non_car_data)
	x = np.vstack((car_features,non_car_features)).astype(np.float64)
	y = np.hstack((np.ones(len(car_features)),np.zeros(len(non_car_features))))
	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)
	return X_train,X_test,y_train,y_test

def train_model(X_train,y_train):
	svc = LinearSVC()
	svc.fit(X_train,y_train)
	return svc

图像搜索

接下来进行车辆搜索，其原理是使用滑动窗口截取图像部分区域，然后用刚刚训练好的svm模型判断是否是车辆，如果是就保存该区域坐标。

定义滑动窗口

首先定义滑动窗口，其中有这样几个概念cell、block、window，step。cell是抽取hog特征时的最小单位，这里定义为16个像素。block由cell组成，这里定义为2*2个cell。window是滑动窗口大小，定义为64个像素。step是步长，即每次滑动的长度，定义为2个cell。

#x方向上block个数
nxblock = (ch1.shape[1]//pix_per_cell) + 1
#y方向上block个数
nyblock = (ch1.shape[0]//pix_per_cell) + 1
#定义window大小
window = 64
#每个window的cell个数，为了后续计算，剪掉1
nblocks_per_window = (window//pix_per_cell) - 1
#步长
cells_per_step = 2
#x方向上需要移动的次数
nxstep = (nxblock - nblocks_per_window)//cells_per_step
#y方向上需要移动的次数
nystep = (nyblock - nblocks_per_window)//cells_per_step

滑动窗口滑动过程如下

分类

抽取窗口内图像特征，并进行分类，若为车辆，保存其窗口坐标。

#获取特征
hog1 = get_hog_feature(ch1,orient,pix_per_cell,cell_per_block, feature_vec = False)
if hog_channel == 'ALL':
	hog2 = get_hog_feature(ch2,orient,pix_per_cell,cell_per_block, feature_vec = False)
	hog3 = get_hog_feature(ch3,orient,pix_per_cell,cell_per_block, feature_vec = False)
#滑动抽取每个窗口特征
for xb in range(nxstep):
	for yb in range(nystep):
		ypos = yb * cells_per_step
		xpos = xb * cells_per_step
		hog_feat1 = hog1[ypos:ypos+nblocks_per_window,xpos:xpos+nblocks_per_window].ravel()
		if hog_channel == 'ALL':
			hog_feat2 = hog2[ypos:ypos+nblocks_per_window,xpos:xpos+nblocks_per_window].ravel()
			hog_feat3 = hog3[ypos:ypos+nblocks_per_window,xpos:xpos+nblocks_per_window].ravel()
			hog_features = np.hstack((hog_feat1, hog_feat2, hog_feat3))
		else : hog_features = hog_feat1

		xleft = xpos * pix_per_cell
		ytop = ypos * pix_per_cell
		hog_features = hog_features.reshape(1,-1)
		test_prediction = svc.predict(hog_features)
		#若是车辆
		if test_prediction == 1 or show_all_rectangle:
			xbox_left = np.int(xleft * scale)
			ytop = np.int(ytop * scale)
			win = np.int(window * scale)
			#保存窗口坐标
			rectangles.append(((xbox_left,ytop + ystart),(xbox_left + window,ytop+win+ystart)))