Abstract:

I will be presenting two of our recent works on 3D point cloud registration:

1. A scene flow method for non-rigid registration:

I will discuss our current method to recover scene flow from point clouds. Scene flow is the three-dimensional (3D) motion field of a scene, and it provides information about the spatial arrangement and rate of change of objects in dynamic environments. We use the graph Laplacian to regularize the scene flow to be “as-rigid-as-possible,” and our method can be used with or without learning.

2. A rigid registration method based on PointNet and the Lucas-Kanade (LK) algorithm:

I will discuss our recent learning-based point cloud registration method. We revisit a recent innovation—PointNetLK—and show that the inclusion of an analytical Jacobian can exhibit great generalization properties while reaping the inherent fidelity benefits of a learning framework.

Bonus: I will discuss the new Argoverse Stereo dataset for autonomous driving and the Argoverse Stereo Competition.

BIO:

Jhony Kaesemodel Pontes is a research scientist at Argo AI working on computer vision and machine learning for autonomous vehicles. He completed his Ph.D. in 2019 at the Queensland University of Technology in Australia. His research interests are in 3D vision, specifically reconstruction and registration.

Homepage:  https://jhonykaesemodel.com/

Sponsored in part by:   Facebook Reality Labs Pittsburgh