OpenCV像素遍历常用的几种方法
以从 organized 的点云提取 RGB 信息为例
动态地址at
基于Mat对象的随机像素访问 API 实现，通过行列索引方式遍历每个像素值。这种方法速度较慢，不太适合用于像素遍历。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
| void extract_1(const pcl::PointCloud<PointT>::Ptr cloud, cv::Mat &image)
{
auto &cloud_height = cloud->height;
auto &cloud_width = cloud->width;
image = cv::Mat(cv::Size(cloud_width, cloud_height), CV_8UC3);
#pragma omp parallel for
for (size_t row = 0; row < cloud_height; row++)
{
for (size_t col = 0; col < cloud_width; col++)
{
auto index = row * cloud_width + col;
const auto &pt = cloud->points[index];
image.at<cv::Vec3b>(row, col)[0] = pt.b;
image.at<cv::Vec3b>(row, col)[1] = pt.g;
image.at<cv::Vec3b>(row, col)[2] = pt.r;
}
}
}
|
行指针
基于Mat对象的行随机访问指针方式实现对每个像素的遍历。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
| void extract_2(const pcl::PointCloud<PointT>::Ptr cloud, cv::Mat &image)
{
auto &cloud_height = cloud->height;
auto &cloud_width = cloud->width;
image = cv::Mat(cv::Size(cloud_width, cloud_height), CV_8UC3);
#pragma omp parallel for
for (size_t row = 0; row < cloud_height; row++)
{
cv::Vec3b *ptr = image.ptr<cv::Vec3b>(row);
for (size_t col = 0; col < cloud_width; col++)
{
auto index = row * cloud_width + col;
const auto &pt = cloud->points[index];
cv::Vec3b &pixel = ptr[col];
pixel[0] = pt.b;
pixel[1] = pt.g;
pixel[2] = pt.r;
}
}
}
|
uchar 行指针
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
| void extract_3(const pcl::PointCloud<PointT>::Ptr cloud, cv::Mat &image)
{
auto &cloud_height = cloud->height;
auto &cloud_width = cloud->width;
image = cv::Mat(cv::Size(cloud_width, cloud_height), CV_8UC3);
#pragma omp parallel for
for (size_t row = 0; row < cloud_height; row++)
{
uchar *uc_pixel = image.ptr(row);
for (size_t col = 0; col < cloud_width; col++)
{
auto index = row * cloud_width + col;
const auto &pt = cloud->points[index];
uc_pixel[0] = pt.b;
uc_pixel[1] = pt.g;
uc_pixel[2] = pt.r;
uc_pixel += 3;
}
}
}
|
将图像完全展开
一般图像行与行之间往往存储是不连续的，但是有些图像可以是连续的，Mat提供了一个检测图像是否连续的函数isContinuous()。当图像连通时，我们就可以把图像完全展开，看成是一行进行处理。
速度比较
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
| int main()
{
std::string cloud_path = "test.pcd"
pcl::PointCloud<PointT>::Ptr cloud(new pcl::PointCloud<PointT>);
pcl_utils::load_pcd(cloud_path, *cloud);
std::cout << *cloud << std::endl;
auto start = std::chrono::steady_clock::now();
cv::Mat img_1;
extract_1(cloud, img_1);
auto end = std::chrono::steady_clock::now();
auto elapsed_seconds = static_cast<std::chrono::duration<double>>(end - start);
std::cout << "1 elapsed time: " << elapsed_seconds.count() << "s\n";
start = std::chrono::steady_clock::now();
cv::Mat img_2;
extract_2(cloud, img_2);
end = std::chrono::steady_clock::now();
elapsed_seconds = static_cast<std::chrono::duration<double>>(end - start);
std::cout << "2 elapsed time: " << elapsed_seconds.count() << "s\n";
start = std::chrono::steady_clock::now();
cv::Mat img_3;
extract_3(cloud, img_3);
end = std::chrono::steady_clock::now();
elapsed_seconds = static_cast<std::chrono::duration<double>>(end - start);
std::cout << "3 elapsed time: " << elapsed_seconds.count() << "s\n";
return 0;
}
|
输出 :
1
2
3
4
5
6
7
8
9
10
11
12
13
| Load pcd file with 1228800 data points.
header: seq: 0 stamp: 0 frame_id:
points[]: 1228800
width: 1280
height: 960
is_dense: 0
sensor origin (xyz): [0, 0, 0] / orientation (xyzw): [0, 0, 0, 1]
1 elapsed time: 0.0199077s
2 elapsed time: 0.012727s
3 elapsed time: 0.00880297s
4 elapsed time: 0.00900436s
|