Publish D435/D455 realsense raw image without realsense-ros package. Test environment Ubuntu18.04/20.04 LTS, ROS melodic, neotic

Dependency

Opencv (Should have if you have installed ROS)
libRealsense2
Eigen (Should have if you have installed ROS)

Install

cd your_ros_workspace/src
git clone https://github.com/arpg/realsense2-lite.git
cd ..
catkin_make

Use

Connect your realsense D435, open a terminal

roscore

In another terminal

cd your_ros_workspace
source devel/setup.bash
roslaunch rs2_lite pub_realsense_image.launch

You should be able to get 4 topics from ir sensor, depth, rgb, rgb_raw. The rgb topic is aligned with the /image0. You need tp let the publish_aligned_rgb to be True to have this topic. rgb_raw is the original rgb image. Align the rgb with depth causes a delay.

/image0
/image1
/depth
/rgb
/rgb_raw

The image width is 640X480 and frquency is 30 Hz, you can modify them in the code. But the realsense can only work in certain frame frequency and image size, the default setting is recommended.

Use the received message

The image topic you receive is encoded in ros message type cv_bridge::CvImage.
For ir sensor the image encoding is mono8, depth is 32FC1 and rgb is rgb8. If you want to extract the opencv Mat after you receive image. Your subcriber's call back function for ir sensor should be something like

void img0_callback(const sensor_msgs::ImageConstPtr &img_msg)
{
    cv_bridge::CvImageConstPtr ptr;

    sensor_msgs::Image img;
    img.header = img_msg->header;
    img.height = img_msg->height;
    img.width = img_msg->width;
    img.is_bigendian = img_msg->is_bigendian;
    img.step = img_msg->step;
    img.data = img_msg->data;
    img.encoding = "mono8";
    ptr = cv_bridge::toCvCopy(img, sensor_msgs::image_encodings::MONO8);

    cv::Mat img = ptr->image.clone();
    return img;
}

for depth image it should be something like

void depth_input(const sensor_msgs::ImageConstPtr &depth_input)
{    
    cv_bridge::CvImagePtr image_ptr;
    image_ptr = cv_bridge::toCvCopy(depth_input, depth_input->encoding);
    constexpr double kDepthScalingFactor = 0.001;//0.001 for realsense, which means 1000 depth means 1 meters
    if(depth_input->encoding == sensor_msgs::image_encodings::TYPE_16UC1)
        (image_ptr->image).convertTo(image_ptr->image, CV_32FC1, kDepthScalingFactor);
    cv::Mat img = image_ptr->image;
    ros::Time stamp = image_ptr->header.stamp;
}

After your callback function you should be able to get an opencv image img