Selectin 3D object recognition consists of a spatial processing unit and a temporal processing unit. The spatial processor determines the type and pose of the object on a per image basis, and the temporal processor uses a track centric multiple hypothesis tracker (MHT) to track objects over time.

The spatial processing unit can be further broken down into three stages: The segmentation processor, the Initial Type, Pose, Geometry (ITPG) estimator, and the Refined Type Pose Geometry (RTPG) estimator. This ITPG returns several initial estimates of the target state using a subspace (Hyperspherical Harmonic) method and the RTPG refines these estimates using fast GPU based correlations and a non-linear optimization algorithm.

For a more detailed discussion of the Selectin process and software architecture, please download the Selectin Whitepaper.