图片红外行人检测，利用python，ROI检测法，contor saliency map ，pedestrian segmentation，SDH feature，SDH feature，sparse representation classifier，并给出代码

以下是基于python的图片红外行人检测的代码：

import cv2
import numpy as np
from sklearn.linear_model import OrthogonalMatchingPursuit

# 读取图片
img = cv2.imread('image.jpg')

# 将图像转换为灰度图像
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# ROI检测
roi = gray[200:600, 200:800]

# Contor Saliency Map提取
blur = cv2.GaussianBlur(roi, (5, 5), 0)
canny = cv2.Canny(blur, 30, 100)
contours, hierarchy = cv2.findContours(canny, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cv2.drawContours(roi, contours, -1, (0, 255, 0), 2)

# 行人分割
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
dilation = cv2.dilate(roi, kernel, iterations=2)
erosion = cv2.erode(dilation, kernel, iterations=2)
ret, binary = cv2.threshold(erosion, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)

# SDH特征提取
patch_size = 16
stride = 8
scales = [1, 2, 4]
features = []
for scale in scales:
    scaled = cv2.resize(binary, (int(binary.shape[1] / scale), int(binary.shape[0] / scale)), interpolation=cv2.INTER_CUBIC)
    for i in range(0, scaled.shape[0] - patch_size, stride):
        for j in range(0, scaled.shape[1] - patch_size, stride):
            patch = scaled[i:i + patch_size, j:j + patch_size]
            features.append(patch.reshape(-1))

features = np.array(features)

# 稀疏表示分类器
model = OrthogonalMatchingPursuit(n_nonzero_coefs=3)
model.fit(features.T, np.zeros(features.shape[0]))

# 标记行人位置
for i in range(features.shape[0]):
    if model.coef_[i] != 0:
        i *= stride
        j *= stride
        x1 = int(j * scales[i // patch_size] + 200)
        y1 = int(i * scales[i // patch_size] + 200)
        x2 = int((j + patch_size) * scales[i // patch_size] + 200)
        y2 = int((i + patch_size) * scales[i // patch_size] + 200)
        cv2.rectangle(img, (x1, y1), (x2, y2), (0, 0, 255), 2)

# 显示结果
cv2.imshow('result', img)
cv2.waitKey(0)
cv2.destroyAllWindows()

以上代码中，将图像转换为灰度图像后，采用ROI检测法提取感兴趣区域，然后使用Contor Saliency Map提取行人轮廓。接着，采用行人分割的方法得到二值图像，并提取SDH特征。最后，使用稀疏表示分类器标记行人位置。

需要注意的是，以上代码仅供参考，实际应用中可能需要根据具体情况进行调整和优化。