dinghongkai/kinect_azure_programming

kinect_azure_programming

record rosbag

# record all the topics（not recommended, the data volume is large）
rosbag record -O xxx.bag -a

# record the specified topic
rosbag record -O xxx.bag /depth_to_rgb/image_raw /rgb/image_raw

parse rosbag

1. put the directory save_rgbd_from_k4a into ~/catkin_ws/src, then compile catkin_make;
2. run roscore to start ROS_MASTER server;
3. run run.sh $path to create directory and .txt files and start save_rgbd_from_k4a node (for example, we set the path as ~/001 here)

bash run.sh ~/001

this node extracts the timestamp, rgb image & depth image from the rosbag file, saves the info into .txt file

4. then, run the rosbag play command

# -r: specify the play speed
rosbag play xxx.bag -r 0.5

Note that converting CV_16UC1 to CV_32FC1 requires to divide by 1000 to keep the accuracy.

https://github.com/microsoft/Azure_Kinect_ROS_Driver/blob/melodic/src/k4a_ros_device.cpp

k4a_result_t K4AROSDevice::renderDepthToROS(sensor_msgs::ImagePtr& depth_image, k4a::image& k4a_depth_frame)
{
    cv::Mat depth_frame_buffer_mat(k4a_depth_frame.get_height_pixels(), k4a_depth_frame.get_width_pixels(), CV_16UC1, k4a_depth_frame.get_buffer());
    cv::Mat new_image(k4a_depth_frame.get_height_pixels(), k4a_depth_frame.get_width_pixels(), CV_32FC1);

    depth_frame_buffer_mat.convertTo(new_image, CV_32FC1, 1.0 / 1000.0f);

    depth_image = cv_bridge::CvImage(std_msgs::Header(), sensor_msgs::image_encodings::TYPE_32FC1, new_image).toImageMsg();

  return K4A_RESULT_SUCCEEDED;
}

data association

associate the rgb image and depth image according to the timestamp, generate the associate.txt file, find the corresponding rgb & depth, rename and copy them into the matched folder

python3 associate.py ~/001

Note: this is modified according to TUM RGB-D Benchmark associate.py