kunzel/strands-object-classification

C++

strands-object-classification

A object recognition framework based on learned object classes.

Prerequisites

PCL

Installation of PCL

Get the PCL source code:

 git clone https://github.com/PointCloudLibrary/pcl pcl-trunk

Configure and compile PCL:

NOTE: when compiling PCL please make sure that "opt/ros/groovy/setup.bash" is NOT sourced (see http://www.pcl-users.org/PCL-1-7-Compilation-Problems-td4028868.html)
```
 cd pcl-trunk && mkdir build && cd build
 ccmake ..
```
In the interactive menu of ccmake hit c for configure. Then enable the variable BUILD_apps (by setting it to ON) and hit c for configure. Afterward set the varibale BUILD_app_3d_rec_framework to ON and configure again (hit c). Finally, hit g for generate and exit.

Back on the command line type:
```
 make
```
The installation of PCL is optional, as we can build the ROS package against the library in PCL's build directory:
```
 sudo make install 
```

Installation of this package

Get the code:

 git clone https://github.com/arbeitor/strands-object-classification

Compile the code using catkin (Please note: it will fail! Just continue with the next step):

 cd strands-object-classification
 catkin_make --force-cmake -DPCL_DIR=/path/tp/pcl/build

As the compilation above fails, configure now two missing variables: PCL_APP_3D_REC_FRAMEWORK_INCLUDE_DIR and PCL_APP_3D_REC_FRAMEWORK_LIBRARY. Do this as follows:
```
 cd build
 ccmake ../src -DCATKIN_DEVEL_PREFIX=../devel -DPCL_DIR=/path/to/pcl/build
```
Within the interactive menu of ccmake set the variables:

PCL_APP_3D_REC_FRAMEWORK_INCLUDE_DIR to /path/to/pcl/pcl/apps/3d_rec_framework/include and

PCL_APP_3D_REC_FRAMEWORK_LIBRARY to �/path/to/pcl/build/lib/libpcl_3d_rec_framework.so

Finally, configure (press c), and generate and exit (press g).

Compile again:

 cd ..
 catkin_make --force-cmake -DPCL_DIR=/path/tp/pcl/build

Getting Started

Get a models archive and unzip somewhere on your disk.

Start some terminals and run the commands below:

Fire up roscore:
```
 $ roscore
```
Start the recognition service (The duration of the training phase depends on how many models you use):
```
 $ rosrun shape_simple_classifier shape_simple_classifier_node -models_dir /path/to/models/data/ -training_dir /path/to/models/trained/ -nn 10
```
Use -chop_z to cut off all information beyond a given distance. FOr example: -chop_z 2

Plug-in the kinect and start openni:

 $ roslaunch openni_launch openni.launch

Run a client to test the service:
```
 $ rosrun soc_test soc_test_node
```