Parts-based 3D Object Pose Estimation
The task of 3D object pose estimation consists of locating and orienting an object in 3D space. Many solutions to this problem make use of a complex representation of the object, such as 3D CAD models or point clouds. Unfortunately, this can prove to be unmanageable in real-world settings due to the lack of such high-fidelity representations or due to the growing size of the object catalog. Inspired by recent advancements in 3D object decomposition, we present a method for 3D object pose estimation that instead uses a compact parametric representation. Using this simple representation as a prerequisite, our method first predicts the pose of the parts of the object, then combines them into a final pose estimation. We demonstrate the success of our parts-based method by comparing its performance to that of a standard baseline method.
[This project was part of my good friend, Ziad Ben Hadj-Alouane ’s master’s thesis at UPenn]