HOG+SVM进行行人检测（github代码解读）

HOG+SVM进行行人检测

代码参考+数据集来源：https://github.com/FrankMa123/-

注意使用的scikit-learn库的版本为0.22，高于该版本的scikit-learn库，代码会报出错误（joblib库无法正常使用）

本文：主要对原作者的代码进行注释，以便于理解HOG+SVM检测，并有利于将该模型应用到其他分类。

致敬原作者！！！

一、config.py

#包含人的图片数据集的路径
pos_im_path = '../data/images/pos_person'
#不包含人的图片数据集的路径
neg_im_path = '../data/images/neg_person'
min_wdw_sz = [68, 124]
step_size = [10, 10]
orientations = 9
#一个cell为6*6个像素点
pixels_per_cell = [6, 6]
#一个block包含2*2个cell
cells_per_block = [2, 2]
visualize = False
normalize = True
pos_feat_ph = '../data/features/pos'
neg_feat_ph = '../data/features/neg'
model_path = '../data/models/'
threshold = .3

二、extract_features.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Sat Apr 20 14:04:30 2019

@author: maqianli
"""

from skimage.feature import hog
from skimage.io import imread
import joblib
import glob
import os
from config import *

def extract_features():
    if not os.path.isdir(pos_feat_ph):
        os.makedirs(pos_feat_ph)
    if not os.path.isdir(neg_feat_ph):
        os.makedirs(neg_feat_ph)
        
    for im_path in glob.glob(os.path.join(pos_im_path, "*")):
        #对每一张包含人的图片进行处理
        im = imread(im_path, as_grey=True)
        print(im.shape)
        #求取hog特征
        fd = hog(im, orientations=9, pixels_per_cell=[8,8], cells_per_block=[2,2], 
                 visualise=False, transform_sqrt=True)
        print(fd.shape)
        #将得到的特征进行存储
        fd_name = os.path.split(im_path)[1].split(".")[0] + ".feat"
        fd_path = os.path.join(pos_feat_ph, fd_name)
        joblib.dump(fd, fd_path)
       
    for im_path in glob.glob(os.path.join(neg_im_path, "*")):
        #对每一张不包含人的图片进行处理
        im = imread(im_path, as_grey=True)
        #求取hog特征
        fd = hog(im, orientations=9, pixels_per_cell=[8,8], cells_per_block=[2,2], 
                 visualise=False, transform_sqrt=True)
        #将得到的特征进行存储
        fd_name = os.path.split(im_path)[1].split(".")[0] + ".feat"
        fd_path = os.path.join(neg_feat_ph, fd_name)
        joblib.dump(fd, fd_path)
    
    print('completed')

if __name__=='__main__':
    extract_features()

三、train_svm.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Sat Apr 20 14:56:42 2019

@author: maqianli
"""

#以上信息为原作者信息

from sklearn.svm import LinearSVC
import joblib
import glob
import os
from config import *
import numpy as np


def train_svm():
    #利用hog特征训练SVM
    pos_feat_path = '../data/features/pos'
    neg_feat_path = '../data/features/neg'
    model_path = '../data/models/svm_model'
    #分别存放特征和标签
    fds, labels = [], []
    #对得到的特征进行打标签，分别包含有人和无人的情况
    for feat_path in glob.glob(os.path.join(pos_feat_path, "*.feat")):
        fd = joblib.load(feat_path)
        fds.append(fd)
        labels.append(1)
    
    for feat_path in glob.glob(os.path.join(neg_feat_path, "*.feat")):
        fd = joblib.load(feat_path)
        fds.append(fd)
        labels.append(0)
    #用线性SVM
    clf = LinearSVC()
    clf.fit(fds, labels)
    
    if not os.path.isdir(os.path.split(model_path)[0]):
        os.makedirs(os.path.split(model_path)[0])
    #print('Classifier save to {}'.format(model_path))
    
    #将训练好的模型存储到model_path下
    joblib.dump(clf, model_path)
    print('Classifier save to {}'.format(model_path))
    
if __name__=='__main__':
    train_svm()

四、detector.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Sat Apr 20 15:33:30 2019

@author: maqianli
"""

import numpy as np
from skimage.transform import pyramid_gaussian
from imutils.object_detection import non_max_suppression
import imutils
from skimage.feature import hog
import joblib
import cv2
from config import *
from skimage import color
import matplotlib.pyplot as plt
import os
import glob

model_path = '../data/models/svm_model'


#滑动窗口，即得到图片的一部分
def sliding_window(image, window_size, step_size):
    for y in range(0, image.shape[0], step_size[1]):
        for x in range(0, image.shape[1], step_size[0]):
            yield x, y, image[y: y + window_size[1], x:x + window_size[0]]


def detector(filename):
    im = cv2.imread(filename)
    #对图片大小进行控制
    im = imutils.resize(im, width=min(400, im.shape[1]))
    #最小窗口大小
    min_wdw_sz = (64, 128)
    step_size = (10, 10)
    #图像金字塔缩放倍数
    downscale = 1.25
    #导入训练好的SVM模型
    clf = joblib.load(model_path)
    #detections存放图片中的人物
    detections = []
    #scale为缩放的次数
    scale = 0
    #对图片进行缩放
    for im_scaled in pyramid_gaussian(im, downscale=downscale):
        #如果图片比我们规定的图片小，那么就结束，不进行下面的hog检测
        if im_scaled.shape[0] < min_wdw_sz[1] or im_scaled.shape[1] < min_wdw_sz[0]:
            break
        #对滑动窗口进行hog特征检测
        for (x, y, im_window) in sliding_window(im_scaled, min_wdw_sz, step_size):
            if im_window.shape[0] != min_wdw_sz[1] or im_window.shape[1] != min_wdw_sz[0]:
                continue
            #将得到的图像进行灰度化
            im_window = color.rgb2gray(im_window)
            #hog特征检测
            fd = hog(im_window, orientations=9, pixels_per_cell=[8, 8], cells_per_block=[2, 2],
                     transform_sqrt=True)
			#将fd特征转换成向量，并调用训练好的模型进行预测
            fd = fd.reshape(1, -1)
            pred = clf.predict(fd)
			
            #如果检测到图片中有人
            if pred == 1:
                if clf.decision_function(fd) > 0.5:
                    #保存 包含人的矩形
                    detections.append((int(x * (downscale ** scale)), int(y * (downscale ** scale)),
                                       clf.decision_function(fd),
                                       int(min_wdw_sz[0] * (downscale ** scale)),
                                       int(min_wdw_sz[1] * (downscale ** scale))))

        scale += 1

    #克隆一遍原图，画出矩形，进行输出
    clone = im.copy()
    for (x_t1, y_t1, _, w, h) in detections:
        cv2.rectangle(im, (x_t1, y_t1), (x_t1 + w, y_t1 + h), (0, 255, 0), thickness=2)

    rects = np.array([[x, y, x + w, y + h] for (x, y, _, w, h) in detections])
    sc = [score[0] for (x, y, score, w, h) in detections]
    print("sc: ", sc)
    sc = np.array(sc)
    pick = non_max_suppression(rects, probs=sc, overlapThresh=0.3)
    # print("shape, ",pick.shape)

    for (xA, yA, xB, yB) in pick:
        cv2.rectangle(clone, (xA, yA), (xB, yB), (0, 255, 0), 2)

    plt.axis("off")
    plt.imshow(cv2.cvtColor(im, cv2.COLOR_BGR2RGB))
    plt.title('Raw Detection before NMS')
    plt.show()

    plt.axis("off")
    plt.imshow(cv2.cvtColor(clone, cv2.COLOR_BGR2RGB))
    plt.title("Final Detections after applying NMS")
    plt.show()


def test_folder(foldername):
    filenames = glob.iglob(os.path.join(foldername, '*'))
    for filename in filenames:
        #对foldername下的每一张图片进行detect
        print(filename)
        detector(filename)


if __name__ == '__main__':
    #对test_image里面的图片进行预测
    foldername = 'test_image'
    test_folder(foldername)