Object Analytics (OA) is ROS2 wrapper for realtime object tracking and 3D localization. These packages aim to provide real-time object analyses over RGB-D camera inputs, enabling ROS developer to easily create amazing robotics advanced features, like intelligent collision avoidance, people follow and semantic SLAM. It consumes sensor_msgs::PointClould2 data delivered by RGB-D camera, subscribs topic on object detection by ros2_intel_movidius_ncs, publishs topics on object tracking in 2D RGB image and object localization in 3D camera coordination system.
OA keeps integrating with various "state-of-the-art" algorithms.
- Object detection offload to VPU, Intel Movidius NCS, with MobileNet SSD model and Caffe framework(TODO).
We support Ubuntu Linux Bionic Beaver 18.04 on 64-bit. We not support Mac OS X and Windows.
- Intel NUC (CPU: Intel i7-6700HQ @2.60GHz, Mem:16G)
- Intel Movidius Neural Compute Stick
- Intel RealSense D435/D415
Install ROS2 desktop packages ros-crystal-desktop
sudo apt-get install ros-crystal-desktop
The ros-crystal-desktop will include below packages.
- ament_cmake
- std_msgs
- sensor_msgs
- geometry_msgs
- rclcpp
- rosidl_default_generators
- rosidl_interface_packages
- launch
- ros2run
- class_loader
- pcl_conversions
sudo apt-get install ros-crystal-cv-bridge ros-crystal-object-msgs ros-crystal-image-transport ros-crystal-librealsense2 ros-crystal-realsense-camera-msgs ros-crystal-realsense-ros2-camera
- cv_bridge
- object_msgs
- ros2_message_filters
- ros2_intel_realsense (The only supported RGB-D camera by now is Intel RealSense)
ros2_intel_movidius has not integrated in ROS2 release, so there is no debian package available for Movidius NCS installation, need to build from source, more details please referece to https://github.com/intel/ros2_intel_movidius_ncs).
# build ncsdk
mkdir ~/code
cd ~/code
git clone https://github.com/movidius/ncsdk
git clone https://github.com/movidius/ncappzoo
cd ~/code/ncsdk
make install
ln -sf ~/code/ncappzoo /opt/movidius/ncappzoo
# build ros2_intel_movidius_ncs
mkdir ~/ros2_ws/src -p
cd ~/ros2_ws/src
git clone https://github.com/intel/ros2_intel_movidius_ncs.git
cd ~/ros2_ws
source /opt/ros/crystal/setup.bash
colcon build --symlink-install (Install python3-colcon-common-extensions by apt-get if colcon command not exist)
# build CNN model (Please plugin NCS device on the host while compiling)
cd /opt/movidius/ncappzoo/caffe/SSD_MobileNet
make
# Copy object label file to NCSDK installation location.
cp ~/ros2_ws/src/ros2_intel_movidius_ncs/data/labels/* /opt/movidius/ncappzoo/data/ilsvrc12/
sudo apt-get install ros-crystal-object-analytics-msgs ros-crystal-object-analytics-node ros-crystal-object-analytics-rviz
The object analytics packages installation have been completed. You could jump to Run for executing, you could also install OA from source for more features. Notes: debian installed package does not support 2d tracking feature as the dependent opencv3.3 has no debian available. For full feature, please build opencv3.3 and install object analytics from source.
- OpenCV3 & opencv-contrib 3.3 (OA depends on tracking feature from OpenCV Contrib 3.3. OpenCV 3.3 is not integrated in ROS2 Crystal release, need to build and install Opencv3 with contrib from source to apply tracking feature)
# Build and Install OpenCV3 with opencv-contrib
mkdir ${HOME}/opencv
cd ${HOME}/opencv
git clone https://github.com/opencv/opencv.git -b 3.3.0
git clone https://github.com/opencv/opencv_contrib.git -b 3.3.0
mkdir opencv/build -p
cd opencv/build
cmake -DOPENCV_EXTRA_MODULES_PATH=${HOME}/opencv/opencv_contrib/modules \
-DCMAKE_INSTALL_PREFIX=/usr/local -DBUILD_opencv_cnn_3dobj=OFF ..
make -j8
sudo make install
sudo ldconfig
# get code
mkdir ~/ros2_ws/src -p
cd ~/ros2_ws/src
git clone https://github.com/intel/ros2_object_analytics.git -b devel (devel branch is the latest code with 2D tracking features, while master branch is stable for ros2 bloom release)
# Build
cd ~/ros2_ws
source /opt/ros/crystal/setup.bash
colcon build --symlink-installsource /opt/ros/crystal/setup.bash
source ~/ros2_ws/install/local_setup.bash
echo -e "param_file: mobilenetssd.yaml\ninput_topic: /object_analytics/rgb" > `ros2 pkg prefix movidius_ncs_launch`/share/movidius_ncs_launch/config/default.yaml
source /opt/ros/crystal/setup.bash
source ~/ros2_ws/install/local_setup.bash
ros2 launch object_analytics_node object_analytics.launch.py
/object_analytics/detected_objects (object_msgs::msg::ObjectsInBoxes)
/object_analytics/rgb (sensor_msgs::msg::Image)
/object_analytics/pointcloud (sensor_msgs::msg::PointCloud2)
/object_analytics/localization (object_analytics_msgs::msg::ObjectsInBoxes3D)
/object_analytics/tracking (object_analytics_msgs::msg::TrackedObjects)
/object_analytics/movement (object_analytics_msgs::msg::MovingObjectsInFrame)
By default, object analytics will launch both tracking and localization features, but either tracking or localization or both can be dropped. Detailed please refer comments embedded in launch file.
To ensure the algorims in OA components to archive best performance in ROS2, we have below tools used to examine design/development performance/accuracy/precision..., more tools are in developing progress and will publish later.
The tools is used to feed tracking node with raw images from datasets within fixed time interval(33ms), also simulate detector send groundtruth as detections to tracking node for rectification, then receive tracking results for precision and recall stastics. It support multiple algorithms(dynamic configure to tracking node when start).
# ros2 run object_analytics_node tracker_regression --options
options: [-a algorithm] [-p dataset_path] [-t dataset_type] [-n dataset_name] [-h];
-h : Print this help function.
-a algorithm_name : Specify the tracking algorithm in the tracker.
supported algorithms: KCF,TLD,BOOSTING,MEDIAN_FLOW,MIL,GOTURN.
-p dataset_path : Specify the tracking datasets location.
-t dataset_type : Specify the dataset type: video,image.
-n dataset_name : Specify the dataset name
Video dataset with tracking algorithm("MEDIAN_FLOW"):
# ros2 run object_analytics_node tracker_regression -p /your/video/datasets/root/path -t video -n dudek -a MEDIAN_FLOW
Image dataset with default algorithm("MEDIAN_FLOW"):
# ros2 run object_analytics_node tracker_regression -p /your/image/datasets/root/path -t image -n Biker -a MEDIAN_FLOW
Support both video and image dataset, but you may need to translate into below formats.
Video dataset: (Refer to opencv_extra tracking dataset)
track_vid/ (/your/video/datasets/root/path)
├── david
│ ├── data
│ │ └── david.webm
│ ├── david.yml
│ ├── gt.txt
│ └── initOmit
│ └── david.txt
├── dudek
│ ├── data
│ │ └── dudek.webm
│ ├── dudek.yml
│ ├── gt.txt
│ └── initOmit
│ └── dudek.txt
├── faceocc2
│ ├── data
│ │ └── faceocc2.webm
│ ├── faceocc2.yml
│ ├── gt.txt
│ └── initOmit
│ └── faceocc2.txt
├── list.txt (Note: this is mannually added, list the dataset names which will be used)
└── README.md
Image dataset: (Refer to database from Computer Vision Lab@HYU)
track_img/ (/your/video/datasets/root/path)
├── Biker
├── Bird1
├── Bird2
├── list.txt (Note: this is mannually added, list the dataset names which will be used)
├── Man
├── Matrix
└── Woman