python+opencv3.4.0 实现HOG+SVM行人检测的示例代码_Python

参照opencv官网例程写了一个基于python的行人检测程序，实现了和自带检测器基本一致的检测效果。

网址：https://docs.opencv.org/3.4.0/d5/d77/train_HOG_8cpp-example.html

opencv版本：3.4.0

训练集和opencv官方用了同一个，可以从http://pascal.inrialpes.fr/data/human/下载，在网页的最下方“here（970MB处）”，用迅雷下载比较快（500kB/s）。训练集文件比较乱，需要仔细阅读下载首页的文字介绍。注意pos文件夹下的png图片属性，它们用opencv无法直接打开，linux系统下也无法显示，需要用matlab读取图片->保存才行，很奇怪的操作。

代码如下，尽可能与opencv官方例程保持一致，但省略了很多不是很关键的东西。训练一次大概需要十几分钟

				?

									import cv2

									import numpy as np

									import random

									def load_images(dirname, amout = 9999):

									 img_list = []

									 file = open(dirname)

									 img_name = file.readline()

									 while img_name != '': # 文件尾

									  img_name = dirname.rsplit(r'/', 1)[0] + r'/' + img_name.split('/', 1)[1].strip('\n')

									  img_list.append(cv2.imread(img_name))

									  img_name = file.readline()

									  amout -= 1

									  if amout <= 0: # 控制读取图片的数量

									   break

									 return img_list

									# 从每一张没有人的原始图片中随机裁出10张64*128的图片作为负样本

									def sample_neg(full_neg_lst, neg_list, size):

									 random.seed(1)

									 width, height = size[1], size[0]

									 for i in range(len(full_neg_lst)):

									  for j in range(10):

									   y = int(random.random() * (len(full_neg_lst[i]) - height))

									   x = int(random.random() * (len(full_neg_lst[i][0]) - width))

									   neg_list.append(full_neg_lst[i][y:y + height, x:x + width])

									 return neg_list

									# wsize: 处理图片大小，通常64*128; 输入图片尺寸>= wsize

									def computeHOGs(img_lst, gradient_lst, wsize=(128, 64)):

									 hog = cv2.HOGDescriptor()

									 # hog.winSize = wsize

									 for i in range(len(img_lst)):

									  if img_lst[i].shape[1] >= wsize[1] and img_lst[i].shape[0] >= wsize[0]:

									   roi = img_lst[i][(img_lst[i].shape[0] - wsize[0]) // 2: (img_lst[i].shape[0] - wsize[0]) // 2 + wsize[0], \

									     (img_lst[i].shape[1] - wsize[1]) // 2: (img_lst[i].shape[1] - wsize[1]) // 2 + wsize[1]]

									   gray = cv2.cvtColor(roi, cv2.COLOR_BGR2GRAY)

									   gradient_lst.append(hog.compute(gray))

									 # return gradient_lst

									def get_svm_detector(svm):

									 sv = svm.getSupportVectors()

									 rho, _, _ = svm.getDecisionFunction(0)

									 sv = np.transpose(sv)

									 return np.append(sv, [[-rho]], 0)

									# 主程序

									# 第一步：计算HOG特征

									neg_list = []

									pos_list = []

									gradient_lst = []

									labels = []

									hard_neg_list = []

									svm = cv2.ml.SVM_create()

									pos_list = load_images(r'G:/python_project/INRIAPerson/96X160H96/Train/pos.lst')

									full_neg_lst = load_images(r'G:/python_project/INRIAPerson/train_64x128_H96/neg.lst')

									sample_neg(full_neg_lst, neg_list, [128, 64])

									print(len(neg_list))

									computeHOGs(pos_list, gradient_lst)

									[labels.append(+1) for _ in range(len(pos_list))]

									computeHOGs(neg_list, gradient_lst)

									[labels.append(-1) for _ in range(len(neg_list))]

									# 第二步：训练SVM

									svm.setCoef0(0)

									svm.setCoef0(0.0)

									svm.setDegree(3)

									criteria = (cv2.TERM_CRITERIA_MAX_ITER + cv2.TERM_CRITERIA_EPS, 1000, 1e-3)

									svm.setTermCriteria(criteria)

									svm.setGamma(0)

									svm.setKernel(cv2.ml.SVM_LINEAR)

									svm.setNu(0.5)

									svm.setP(0.1) # for EPSILON_SVR, epsilon in loss function?

									svm.setC(0.01) # From paper, soft classifier

									svm.setType(cv2.ml.SVM_EPS_SVR) # C_SVC # EPSILON_SVR # may be also NU_SVR # do regression task

									svm.train(np.array(gradient_lst), cv2.ml.ROW_SAMPLE, np.array(labels))

									# 第三步：加入识别错误的样本，进行第二轮训练

									# 参考 http://masikkk.com/article/SVM-HOG-HardExample/

									hog = cv2.HOGDescriptor()

									hard_neg_list.clear()

									hog.setSVMDetector(get_svm_detector(svm))

									for i in range(len(full_neg_lst)):

									 rects, wei = hog.detectMultiScale(full_neg_lst[i], winStride=(4, 4),padding=(8, 8), scale=1.05)

									 for (x,y,w,h) in rects:

									  hardExample = full_neg_lst[i][y:y+h, x:x+w]

									  hard_neg_list.append(cv2.resize(hardExample,(64,128)))

									computeHOGs(hard_neg_list, gradient_lst)

									[labels.append(-1) for _ in range(len(hard_neg_list))]

									svm.train(np.array(gradient_lst), cv2.ml.ROW_SAMPLE, np.array(labels))

									# 第四步：保存训练结果

									hog.setSVMDetector(get_svm_detector(svm))

									hog.save('myHogDector.bin')

以下是测试代码：

				?

									import cv2

									import numpy as np

									hog = cv2.HOGDescriptor()

									hog.load('myHogDector.bin')

									cap = cv2.VideoCapture(0)

									while True:

									 ok, img = cap.read()

									 rects, wei = hog.detectMultiScale(img, winStride=(4, 4),padding=(8, 8), scale=1.05)

									 for (x, y, w, h) in rects:

									  cv2.rectangle(img, (x, y), (x + w, y + h), (0, 0, 255), 2)

									 cv2.imshow('a', img)

									 if cv2.waitKey(1)&0xff == 27: # esc键

									  break

									cv2.destroyAllWindows()