RGBDSLAMv2-MIVisionX

This is an implementation of RGBDSLAM_V2 that utilizes AMD MIVisionX for feature detection and ROCm OpenCL for offloading computations to Radeon GPUs.

Prerequisites

Linux

• Ubuntu 16.04
• Radeon Open Compute (ROCm)
• AMD MIVisionX
• OpenCV 3.4
• ROS Kinetic

Follow the steps in INSTRUCTIONS_PREREQUISITES.md to install the prerequisites of RGBDSLAM_v2 on ROS Kinetic.

• Forked version of g2o
• Eigen 3.2.10
• PCL 1.8

Build & Install RGBDSLAM_v2

After installing the prerequisites, use these instructions to install and run RGBDSLAM_v2.

• Make a catkin workspace and clone this repo inside the source folder of the workspace
• Set the environment variable OpenCV_DIR to </path/to/opencv/build>
• Use catkin_make to build

mkdir -p catkin_ws/src
cd catkin_ws/src
git clone https://github.com/ICURO-AI-LAB/RGBDSLAMv2-MIVisionX.git
cd ..
export OpenCV_DIR=</path/to/opencv/build>
catkin_make

Running RGBDSLAM_v2

• Pre-recorded data from a ROSBag

source catkin_ws/devel/setup.bash
roslaunch rgbdslam test_settings.launch bagfile_name:=<path/to/rosbag>

• Live

source catkin_ws/devel/setup.bash
roslaunch rgbdslam rgbdslam.launch

Input ROS Topics

• Camera Info: /camera/rgb/camera_info
• RGB Image : /camera/rgb/image_color
• Depth Image: /camera/depth/image

The camera driver must be run using a separate ROS package. For example, Microsoft Kinect One (Kinect v2) can be run using https://github.com/code-iai/iai_kinect2. Once the driver is running, the topics have to be relayed appropriately.

Sample ROS bags

Sample ROS bags for quickly testing the installation can be found here

Docker

Scripts for building and running a docker image is provided in this directory. This can be used to easily install RGBDSLAMv2 without dependency issues.

Prerequisites to build the docker image:

• Ubuntu 16.04/Ubuntu 18.04
• Docker
• Radeon Open Compute (ROCm)

Dockerfile.rgbdslam builds an image with all the prerequisites installed. Follow the instructions below to install RGBDSLAMv2 in the docker image

Note 1: If you are running Ubuntu 18.04, uncomment lines 115-118 in RGBDSLAMv2-MIVisionX/docker/Dockerfile.rgbdslam Note 2: If you are running Ubuntu 16.04, uncomment lines 121-125 in RGBDSLAMv2-MIVisionX/docker/Dockerfile.rgbdslam Note 3: Building the docker image might take a long time as PCL 1.8 is set to be built using -j1. This can be made faster if desired, but is not recommended as the process takes a large amount of memory.

• Build the docker image

cd RGBDSLAMv2-MIVisionX/docker/
./build

This builds a docker image called rgbdslamv2mivisionx

• Run the image as a container

cd RGBDSLAMv2-MIVisionX/docker/
./run

• Once you are in the container, run RGBDSLAMv2

cd ~
source catkin_ws/devel/setup.bash
roslaunch rgbdslam test_settings.launch bagfile_name:=<path/to/rosbag>