libmultisense 7.1.0

LibMultiSense python package wrapped using pybind11

LibMultiSense

LibMultiSense is a C++ and Python library designed to simplify interaction with the MultiSense S family of stereo sensors developed by Carnegie Robotics. It provides a comprehensive, easy-to-use API for capturing and processing stereo sensor data an generating depth images, color images, and 3D point clouds.

Official product page: Carnegie Robotics MultiSense Products

Detailed documentation: LibMultiSense Documentation

LibMultiSense was recently refactored to have a new API. The following examples in the README all assume the user is using the new API. To build with the new API, the following CMake arguments should be set.

-DBUILD_LEGACY_API=0FF

The README for the Legacy API can be found here.

Table of Contents

• Client Networking Prerequisite
• Quickstart Guide
  • Python
  • C++
• Optional Dependencies
  • OpenCV
  • nlohmann_json
  • pybind11
  • googletest
• Installation
  • Linux
    • Python
    • C++
  • MacOS
    • Python
    • C++
  • Windows
    • Python
    • C++
• CMake Project Integration
  • Local Installation
  • Git Submodule
• Documentation
• Support
• Camera Configuration
  • Python
  • C++
• Point Cloud Generation
  • Python
  • C++
• Depth Image Generation
  • Python
  • C++
• Color Image Generation
  • Python
  • C++

Client Networking Prerequisite

The MultiSense comes preconfigured with a static 10.66.171.21 IP address with a /24 subnet. To connect to the MultiSense, a client machine must be updated with an IP address on the 10.66.171 subnet.

Please see the host network configuration for details on how to set a client machine's IP address and MTU.

Quickstart Guide

Below are minimal examples demonstrating basic usage of LibMultiSense to capture rectified images from the camera.

Before running the examples make sure LibMultiSense is installed, and your machines network is properly configured.

Python

import libmultisense as lms
import cv2

channel_config = lms.ChannelConfig()
channel_config.ip_address = "10.66.171.21"
with lms.Channel.create(channel_config) as channel:
    channel.start_streams([lms.DataSource.LEFT_RECTIFIED_RAW, lms.DataSource.RIGHT_RECTIFIED_RAW])

    while True:
        frame = channel.get_next_image_frame()
        if frame:
            for source, image in frame.images.items():
                cv2.imwrite(str(source) + ".png", image.as_array)

C++

#include <MultiSense/MultiSenseChannel.hh>
#include <MultiSense/MultiSenseUtilities.hh>

int main()
{
    const auto channel = lms::Channel::create(lms::Channel::ChannelConfig{"10.66.171.21"});
    channel->start_streams({lms::DataSource::LEFT_RECTIFIED_RAW, lms::DataSource::RIGHT_RECTIFIED_RAW});

    while(true)
    {
        if (const auto image_frame = channel->get_next_image_frame(); image_frame)
        {
            for (const auto &[source, image]: image_frame->images)
            {
                const auto path = std::to_string(image_frame->frame_id) +  "_" +
                                  std::to_string(static_cast<int>(source)) + ".png";
                lms::write_image(image, path);
            }
        }
    }
    return 0;
}

---

Optional Dependencies

OpenCV

LibMultiSense optionally has OpenCV utility functions to make the LibMultiSense client API easier to integrate with existing systems. To build the OpenCV helpers the following CMake argument should be set

-DBUILD_OPENCV=ON

This will require a system installation of OpenCV, or an installation which can be pointed to with CMake's CMAKE_PREFIX_PATH argument

nlohmann json

LibMultiSense optionally uses nlohmann_json for serialization of base LibMultiSense types. To build the nlohmann_json serialization helpers the following CMake argument should be set

-DBUILD_JSON_SERIALIZATION=ON

This will require a system installation of nlohmann_json, or an installation which can be pointed to with CMake's CMAKE_PREFIX_PATH argument

pybind11

LibMultiSense optionally uses pybind11 to generate python bindings for the C++ API. To build the pybind11 python bindings the following CMake argument should be set

-DBUILD_PYTHON_BINDINGS=ON

This will require a system installation of pybind11, or an installation which can be pointed to with CMake's CMAKE_PREFIX_PATH argument

googletest

LibMultiSense optionally uses googletest for unit testing the C++ API. To build the googletest unit tests the following CMake argument should be set

-DBUILD_TESTS=ON

This will require a system installation of googletest, or an installation which can be pointed to with CMake's CMAKE_PREFIX_PATH argument

---

Installation

Linux

Python

Pre-Compiled Wheel

The latest LibMultiSense Release includes PIP wheels which can be installed on a x86 Linux system in lieu of installing from source. Wheels can be installed with pip using the following command:

pip install <path-to-wheel>

You must install the LibMultiSense wheel which matches your installed Python version. You can check your current Python version with python --version.

For example if python --version returned 3.12.3, you would select the released Linux wheel with cp312 in the name.

From Source

LibMultiSense uses pybind11 to generate Python bindings for the base LibMultiSense API. These bindings can be installed via pip into a Python virtual environment or a local Python project.

To install the LibMultiSense Python bindings

> sudo apt install build-essential pybind11-dev nlohmann-json3-dev

> git clone https://github.com/carnegierobotics/LibMultiSense.git
> cd LibMultiSense
> pip install .

C++

LibMultiSense uses CMake for its build system.

To build the standalone LibMultiSense library and demonstration applications.

# Note this only needs to be run once before building
> sudo apt install build-essential nlohmann-json3-dev

> git clone https://github.com/carnegierobotics/LibMultiSense.git
> cd LibMultiSense
> mkdir build
> cd build && cmake -DBUILD_LEGACY_API=OFF -DBUILD_JSON_SERIALIZATION=ON -DCMAKE_INSTALL_PREFIX=../install ..
> make install
> cd ../install

To build the standalone LibMultiSense library without the demonstration applications, set the cmake variable -DMULTISENSE_BUILD_UTILITIES=OFF

---

MacOS

Python

Pre-Compiled Wheel

The latest LibMultiSense Release includes PIP wheels which can be installed on a ARM64 Mac system in lieu of installing from source. Wheels can be installed with pip using the following command:

pip install <path-to-wheel>

You must install the LibMultiSense wheel which matches your installed Python version. You can check your current Python version with python --version.

For example if python --version returned 3.13.2, you would select the released MacOS wheel with cp313 in the name.

From Source

LibMultiSense uses pybind11 to generate Python bindings for the base LibMultiSense API. These bindings can be installed via pip into a Python virtual environment or a local Python project.

To install the LibMultiSense Python bindings

> brew install pybind11 nlohmann-json

> git clone https://github.com/carnegierobotics/LibMultiSense.git
> cd LibMultiSense
> pip install .

C++

LibMultiSense uses CMake for its build system.

To build the standalone LibMultiSense library and demonstration applications.

# Note this only needs to be run once before building
> brew install nlohmann-json

> git clone https://github.com/carnegierobotics/LibMultiSense.git
> cd LibMultiSense
> mkdir build
> cd build && cmake -DBUILD_LEGACY_API=OFF -DBUILD_JSON_SERIALIZATION=ON -DCMAKE_INSTALL_PREFIX=../install ..
> make install
> cd ../install

To build the standalone LibMultiSense library without the demonstration applications, set the cmake variable -DMULTISENSE_BUILD_UTILITIES=OFF

---

Windows

Python

The latest LibMultiSense Release includes PIP wheels which can be installed on a x86 Windows system in lieu of installing from source. Wheels can be installed with pip using the following command in PowerShell:

pip install <path-to-wheel>

You must install the LibMultiSense wheel which matches your installed Python version. You can check your current Python version in PowerShell with python --version. If Python is not installed on your system, you can use the Microsoft Store on Windows 11 to install the correct version of Python.

For example if python --version returned 3.9.13, you would select the released Windows wheel with cp39 in the name.

Pre-Compiled Wheel

From Source

LibMultiSense uses pybind11 to generate Python bindings for the base LibMultiSense API. These bindings can be installed via pip into a Python virtual environment or a local Python project. To ensure Windows has the proper build tools installed, please install Microsoft Visual Studio with the C++ and CMake extensions.

Note you will need to have a version of Python installed on your Windows system. This was tested with Python 3.9 installed via the Microsoft Store on Windows 11.

To install the LibMultiSense Python bindings open a powershell terminal and execute the following commands

> git clone https://github.com/carnegierobotics/LibMultiSense.git
> cd LibMultiSense
> git clone https://github.com/microsoft/vcpkg.git
> ./vcpkg/bootstrap-vcpkg.bat

> $Env:VCPKG_ROOT = ./vcpkg

> pip install .

C++

LibMultiSense uses CMake and vcpkg to build the LibMultiSense library. To ensure Windows has the proper build tools installed, please install Microsoft Visual Studio with the C++ and CMake extensions.

Open a powershell terminal and execute the following commands:

> git clone https://github.com/carnegierobotics/LibMultiSense.git
> cd LibMultiSense
> git clone https://github.com/microsoft/vcpkg.git
> ./vcpkg/bootstrap-vcpkg.bat

> $Env:VCPKG_ROOT = ./vcpkg

> cmake --build build --config Release --target install

---

CMake Project Integration

Integrating LibMultiSense into an existing CMake project is easy. There are two primary methods for integration: a local install on your system, or a submodule clone within your repository

Local Installation

LibMultiSense is installed on your system (i.e. in a location like /opt/multisense)

find_package(MultiSense)
target_link_libraries(<your-library-or-binary> MultiSense)

When running CMake, make sure to specify the location of the LibMultiSense install via -DCMAKE_PREFIX_PATH

Git Submodule

Clone the LibMultiSense repository into the existing project's source tree. In the main CMakeLists.txt file of the project, add the following lines:

 include_directories(LibMultiSense/source/LibMultiSense)
 add_subdirectory(LibMultiSense/source/LibMultiSense)

---

Documentation

Documentation of high-level LibMultiSense concepts can be found here

Doxygen documentation can be built for LibMultisense by running the Doxygen configuration file located in the docs directory

> cd LibMultiSense/docs
> doxygen Doxyfile

HTML and LaTex documentation will be generated in the docs directory.

Usage examples are included in the Doxygen documentation.

---

Support

To report an issue with this library or request a new feature, please use the GitHub issues system

For product support, please see the support section of our website Individual support requests can be created in our support portal

---

Camera Configuration

Camera settings like resolution, exposure, FPS, gain, gamma, and white balance can be configured via the LibMultiSense image::Config

Python

import libmultisense as lms
import cv2

def main(args):
    channel_config = lms.ChannelConfig()
    channel_config.ip_address = "10.66.171.21"

    with lms.Channel.create(channel_config) as channel:
        if not channel:
            print("Invalid channel")
            exit(1)

        config = channel.get_config()
        config.frames_per_second = 10.0
        config.width = 960
        config.height = 600
        config.disparities = lms.MaxDisparities.D256
        config.image_config.auto_exposure_enabled = True
        config.image_config.gamma = 2.2
        if channel.set_config(config) != lms.Status.OK:
            print("Cannot set configuration")
            exit(1)

C++

#include <MultiSense/MultiSenseChannel.hh>
#include <MultiSense/MultiSenseUtilities.hh>

namespace lms = multisense;

int main(int argc, char** argv)
{
    const auto channel = lms::Channel::create(lms::Channel::ChannelConfig{"10.66.171.21"});
    if (!channel)
    {
        std::cerr << "Failed to create channel" << std::endl;;
        return 1;
    }

    auto config = channel->get_config();
    config.frames_per_second = 10.0;
    config.width = 960;
    config.height = 600;
    config.disparities = lms::MultiSenseConfiguration::MaxDisparities::D256;
    config.image_config.auto_exposure_enabled = true;
    config.image_config.gamma = 2.2;
    if (const auto status = channel->set_config(config); status != lms::Status::OK)
    {
        std::cerr << "Cannot set config" << std::endl;
        return 1;
    }

    return 0;
}

---

Point Cloud Generation

Disparity images can be converted to 3D point cloud images using the client API.

The following modified version of the Quickstart example, converts disparity images to 3D point clouds colorized using the left rectified image.

Python

import libmultisense as lms
import cv2

def main(args):
    channel_config = lms.ChannelConfig()
    channel_config.ip_address = "10.66.171.21"

    with lms.Channel.create(channel_config) as channel:
        if not channel:
            print("Invalid channel")
            exit(1)

        if channel.start_streams([lms.DataSource.LEFT_RECTIFIED_RAW, lms.DataSource.LEFT_DISPARITY_RAW]) != lms.Status.OK:
            print("Unable to start streams")
            exit(1)

        while True:
            frame = channel.get_next_image_frame()
            if frame:
                point_cloud = lms.create_gray8_pointcloud(frame,
                                                         args.max_range,
                                                         lms.DataSource.LEFT_RECTIFIED_RAW,
                                                         lms.DataSource.LEFT_DISPARITY_RAW)

                print("Saving pointcloud for frame id: ", frame.frame_id)
                lms.write_pointcloud_ply(point_cloud, str(frame.frame_id) + ".ply")

C++

#include <MultiSense/MultiSenseChannel.hh>
#include <MultiSense/MultiSenseUtilities.hh>

namespace lms = multisense;

int main(int argc, char** argv)
{
    const auto channel = lms::Channel::create(lms::Channel::ChannelConfig{"10.66.171.21"});
    if (!channel)
    {
        std::cerr << "Failed to create channel" << std::endl;;
        return 1;
    }

    if (const auto status = channel->start_streams({lms::DataSource::LEFT_RECTIFIED_RAW,
                                                    lms::DataSource::LEFT_DISPARITY_RAW}); status != lms::Status::OK)
    {
        std::cerr << "Cannot start streams: " << lms::to_string(status) << std::endl;
        return 1;
    }

    while(!done)
    {
        if (const auto image_frame = channel->get_next_image_frame(); image_frame)
        {
            if (const auto point_cloud = lms::create_color_pointcloud<uint8_t>(image_frame.value(),
                                                                               max_range,
                                                                               lms::DataSource::LEFT_RECTIFIED_RAW,
                                                                               lms::DataSource::LEFT_DISPARITY_RAW); point_cloud)
            {
                std::cout << "Saving pointcloud for frame id: " << image_frame->frame_id << std::endl;
                lms::write_pointcloud_ply(point_cloud.value(), std::to_string(image_frame->frame_id) + ".ply");
            }
        }
    }

    return 0;
}

---

Depth Image Generation

Disparity images can be converted to depth images using the client API

The following modified version of the Quickstart example, converts disparity images to openni depth images and saves them to disk using OpenCV.

Python

import libmultisense as lms
import cv2

def main(args):
    channel_config = lms.ChannelConfig()
    channel_config.ip_address = "10.66.171.21"

    with lms.Channel.create(channel_config) as channel:
        if not channel:
            print("Invalid channel")
            exit(1)

        if channel.start_streams([lms.DataSource.LEFT_DISPARITY_RAW]) != lms.Status.OK:
            print("Unable to start streams")
            exit(1)

        while True:
            frame = channel.get_next_image_frame()
            if frame:
                # MONO16 depth images are quantized to 1 mm per 1 pixel value to match the OpenNI standard
                depth_image = lms.create_depth_image(frame, lms.PixelFormat.MONO16, lms.DataSource.LEFT_DISPARITY_RAW, 65535)
                if depth_image:
                    print("Saving depth image for frame id: ", frame.frame_id)
                    cv2.imwrite(str(frame.frame_id) + ".png", depth_image.as_array)

C++

#include <MultiSense/MultiSenseChannel.hh>
#include <MultiSense/MultiSenseUtilities.hh>

namespace lms = multisense;

int main(int argc, char** argv)
{
    const auto channel = lms::Channel::create(lms::Channel::ChannelConfig{"10.66.171.21"});
    if (!channel)
    {
        std::cerr << "Failed to create channel" << std::endl;;
        return 1;
    }

    if (const auto status = channel->start_streams({lms::DataSource::LEFT_DISPARITY_RAW}); status != lms::Status::OK)
    {
        std::cerr << "Cannot start streams: " << lms::to_string(status) << std::endl;
        return 1;
    }

    while(!done)
    {
        if (const auto image_frame = channel->get_next_image_frame(); image_frame)
        {
            //
            // MONO16 depth will be quantized to mm to match OpenNI's depth format
            //
            if (const auto depth_image = lms::create_depth_image(image_frame.value(),
                                                                 lms::Image::PixelFormat::MONO16,
                                                                 lms::DataSource::LEFT_DISPARITY_RAW,
                                                                 65535); depth_image)
            {
                std::cout << "Saving depth image for frame id: " << image_frame->frame_id << std::endl;
                cv::imwrite(std::to_string(image_frame->frame_id) + ".png", depth_image->cv_mat());
            }
        }
    }

    return 0;
}

---

Color Image Generation

Luma and Chroma Aux images can be converted to BGR color images using the client API

The following modified version of the Quickstart example, converts luma and chroma aux images to BGR images and saves them to disk using OpenCV.

Python

import libmultisense as lms
import cv2

def main(args):
    channel_config = lms.ChannelConfig()
    channel_config.ip_address = "10.66.171.21"

    with lms.Channel.create(channel_config) as channel:
        if not channel:
            print("Invalid channel")
            exit(1)

        if channel.start_streams([lms.DataSource.AUX_RAW]) != lms.Status.OK:
            print("Unable to start streams")
            exit(1)

        while True:
            frame = channel.get_next_image_frame()
            if frame:
                bgr = lms.create_bgr_image(frame, lms.DataSource.AUX_RAW)
                if bgr:
                    cv2.imwrite(str(frame.frame_id) + ".png", bgr.as_array)

C++

#include <MultiSense/MultiSenseChannel.hh>
#include <MultiSense/MultiSenseUtilities.hh>

namespace lms = multisense;

int main(int argc, char** argv)
{
    const auto channel = lms::Channel::create(lms::Channel::ChannelConfig{"10.66.171.21"});
    if (!channel)
    {
        std::cerr << "Failed to create channel" << std::endl;;
        return 1;
    }

    if (const auto status = channel->start_streams({lms::DataSource::LEFT_DISPARITY_RAW}); status != lms::Status::OK)
    {
        std::cerr << "Cannot start streams: " << lms::to_string(status) << std::endl;
        return 1;
    }

    while(!done)
    {
        if (const auto image_frame = channel->get_next_image_frame(); image_frame)
        {
            if (const auto bgr = create_bgr_image(frame, source); bgr)
            {
                cv::imwrite(std::to_string(image_frame->frame_id) + ".png", bgr->cv_mat());
            }
        }
    }

    return 0;
}