Object model acquisition

There are two demos to acquire models of 3D objects. The first one uses only a top-view of the objects, that have to be lying on a plane. The second one uses markers to determine the pose of the camera. This enables to rotate the marker board and acquire all the partial views of an object.

Model acquisition of objects lying on a table

  • Launch rgbd-scan-topview. You might want to enable the —highres flag to get better color textures.

The Kinect must be looking at a dominant plane. Hitting “Acquire new models” should compute a 3D model for all the objects on the table. Note that objects that are too close to each other (about 5cm) might get merged into a single one. The models can be saved into individual objectXX.ply files using the @@Save meshes” button. On the right image you will see a reprojection of the models on the color image, along with the estimated volume of each object in mm3.

  • Note: PCL support is required by this demo.

Model acquisition of objects on a marker board

  • Launch rgbd-scan-markers. You might want to enable the —highres flag to get more color details.

You first need to build a markerboard as the one shown in the video below. To do so, print the two pages located in the rgbd-scan-markers/data directory on A4 paper with a 1:1 scale. Then glue these two sheets on a carton board and put the object in the center. You can then acquire a complete point cloud by slowly rotating the board in front of the camera.

You can then save the acquired model and postprocess it in Meshlab.

More details about the algorithm can be found in this book: http://www.amazon.com/Hacking-Kinect-Jeff-Kramer/dp/1430238674