drv_package

Deep Robot Vision system for NEU household service robot

1. Software Prerequisites

1.1 ROS

We run this package in indigo, other versions of ROS may not be supported.

1.2 Caffe

We use Caffe to implement our tracking and searching methods. More specifically, the tracking method (GOTURN) and face detection use original Caffe, while the object detection method (Faster RCNN) use customized Caffe which has been intergrated into py-faster-rcnn/caffe-faster-rcnn. To install original Caffe, refer http://caffe.berkeleyvision.org/installation.html and follow the 'CMake Build' instruction. After installing it, you need set variable Caffe_ROOT by adding a line export Caffe_ROOT=/home/aicrobo/caffe in your ~/.bashrc. As for caffe-faster-rcnn, refer 1.3.

1.3 py-faster-rcnn

This package is not included in the drv_package, you need to get it from https://github.com/rbgirshick/py-faster-rcnn, and install it following official guide (the customized Caffe installation part is basicly the same with installing ordinary Caffe when using cmake). Notice that if you use cuDNN 5.0 or above, git clone https://github.com/NEU-TEAM/py-faster-rcnn.git instead and cmake (or make). After that, you need pay attention to the lines in file $DRV_PACKAGE/drv_recognize/scripts/process.py which tell where can the prototxt and caffemodel of py-faster-rcnn be found. By default, we have faster_rcnn_test.pt in $PY_FASTER_RCNN/models/DRV/ which is the same as the one in $PY_FASTER_RCNN/models/pascal_voc/VGG16/faster_rcnn_alt_opt/. Besides, we have VGG16_faster_rcnn_final.caffemodel in $PY_FASTER_RCNN/models/DRV/ which you can obtain from Caffe Model Zoo by executing ./data/scripts/fetch_imagenet_models.sh in $PY_FASTER_RCNN.

1.4 rosserial

Please refer Arduino IDE Setup for using rosserial. To install it, we basicly only need to do sudo apt-get install ros-indigo-rosserial-arduino and sudo apt-get install ros-indigo-rosserial We use rosserial_arduino to communicate between one Arduino Uno board (may change to Mega 2560 in near future) and an ADXL345 3-axis accelerator. The Arduino controls 2 servos, one for pitching the RGBD-camera (ASUS Xtion Pro or ASTRA) and the other for rotating it. The accelerator measures the pitch and yaw angles of the camera. The .ino file loaded on the Arduino board is provided in folder $DRV_DIR/supplements/arduino_control, you can load it to your board with Arduino IDE http://arduino.cc/en/Main/Software.

1.5 astra_camera or openni_camera

Using astra_camera is recommended. Specificly, You need first clone the repo https://github.com/orbbec/ros_astra_camera.git into your catkin_ws, then do catkin_make --pkg astra_camera -DFILTER=ON in /catkin_ws. Followed by roscd astra_camera && ./scripts/create_udev_rules to create astra udev rule. If you get error: Getting "fatal error: libudev.h: No such file or directory #include <libudev.h>" when compiling the package, do sudo apt-get install libudev-dev. To use astra_camera, you also need clone https://github.com/orbbec/ros_astra_launch. Using openni_camera by installing from source is also supported, just clone https://github.com/ros-drivers/openni2_camera and catkin_make. You can find the calibration files for our devices (Astra S and Xtion Pro Live) in folder $DRV_DIR/supplements/camera_info for reference purpose. When running in ROS, this folder should be put into /home/YOUR_USERNAME/.ros.

1.6 GOTURN

We use GOTURN as our object tracking backbone. GOTURN itself is not necessarily needed to run this program, we only need the trained model tracker.caffemodel and prototext tracker.prototxt in $Caffe_ROOT/models/GOTURN/. You can get the model from https://github.com/davheld/GOTURN. If your path to these files is different from above, you need to modify the path declarations in $DRV_DIR/drv_track/drv_track.cpp.

1.7 NEU_FACE

We provide the face recognition function based on VGG_Face. To perform face recognition, the caffe model (ex. finetune_neu_face.caffemodel) and prototext (ex. neu_face_deploy.prototxt) should in $Caffe_ROOT/models/neu_face/.

2 Hardware

To run searching and tracking modules smoothly, a workstation with at least 4GB of GRAM is necessary. This program has been tested on multi-machine configurations, in which the host computer's CPUs run at frequencies exceeding 2.8 GHz. Low frequencies will lead to a delay in point cloud capturing and communication between the workstation and the host PC.

3. Installation

1. Clone this repository into catkin_ws/src: git clone https://github.com/NEU-TEAM/drv_package.git
2. Add source ~/catkin_ws/devel/setup.bash to your ~/.bashrc.
3. First do catkin_make --pkg drv_msgs to generate header file used by other nodes.
4. Then run catkin_make to make all the rest packages.

4. Usage

1. Run roscore first.
2. If you use astra_camera, run roslaunch drv_brain drv_astra.launch to launch the whole robot vision system. And if you use OpenNI camera such as Xtion Pro, run launch file: roslaunch drv_brain drv_xtion.launch instead. The only difference between the two launch files is the camera_node being launched.
3. If you use DRV on multiple machines, do roslaunch drv_brain drv_host_astra.launch on host machine (which is on the robot), and roslaunch drv_brain drv_workstation.launch on the workstation to control the robot and process data remotely. Just like 1, you can also do roslaunch drv_brain drv_host_xtion.launch to use Xtion instead.
4. Target can be set by setting rosparam as follows: rosparam set /comm/control/target/label bottle, rosparam set /comm/control/target/is_set true,the setting order should exactly like this. Here the 'bottle' refers to the target label and can be changed to 'chair', 'person', etc. FYI, setting target function and more useful functions beyond that can be easily realized with JARVIS, which is an Android app for controlling the NEU household robot. Till now it only supports Android 6.0 and above.
5. With the target set, the system will automatically run in search mode and finds the target in the scene. If some suspected objects were found, the system will call for user input to judge the result and decide whether continuing searching the target or tracking the confirmed target. If the target is confirmed, the system will run in tracking mode in which the location of the target in 3D space will be generated for manipulating the target.
6. By setting the target is_set param to be false, the system will run in wander mode and do nothing.

5. Trouble Shooting

1. If custom message issue occurred when running catkin_make, run catkin_make --pkg drv_msgs --force-cmake first to make the msg header files needed, and then run catkin_make.
2. If you use astra camera and find the point cloud edge is not quite well, first make sure you get the official source code from https://github.com/orbbec/ros_astra_camera.git, clone it to your catkin_ws and try catkin_make --pkg astra_camera -DFILTER=ON.
3. If your astra_camera node doesn't create topic, try using astra.launch in folder ros_astra_launch (get it from https://github.com/orbbec/ros_astra_launch).

Authors

Zhipeng Dong