Efficient Object Categorization with the Surface-Approximation Polynomials Descriptor

Adding rotational robustness to the Surface-Approximation Polynomials descriptor

“…VFH includes the camera axis in the computation of FPFH histograms to establish viewpoint dependent signatures for the trained objects. The recently proposed SAP descriptor [1] instead directly builds a global descriptor without computing local features and produces categorization results superior to the previous descriptors. We will provide a short description of the SAP descriptor in Section III-B.…”

“…The Surface-Approximation Polynomials (SAP) descriptor has been described in detail in [1]. Therefore, we only provide a schematic summary of the algorithm at this place.…”

“…The Surface-Approximation Polynomials (SAP) descriptor has been recently introduced as a global 3D surface descriptor that is well-suited for the task of object categorization with a robot [1]. The SAP descriptor approximates the surface geometry of a single-shot view onto an object with polynomials.…”

“…The categorization system based on the SAP descriptor can determine the category label of unknown objects that are captured with a depth sensing device like a PMD CamCube or a Microsoft Kinect. It has been shown in [1] that the SAP descriptor is robust enough to compensate smaller viewpoint changes up to 15 • in pan and tilt direction. However, roll rotations of the object or camera cannot be handled at all with the basic approach.…”

“…

The Surface-Approximation Polynomials (SAP) descriptor has been shown to be an appropriate global surface descriptor for object categorization tasks in robotic applications [1]. Nevertheless, in the original formulation the SAP descriptor is not invariant against rotations around the camera axis.

…”