ICCV19-Paper-Review

Summaries of ICCV 2019 papers.

DeepVCP: An End-to-End Deep Neural Network for Point Cloud Registration

Youtube Video

DeepVCP(Virtual Corresponding Points) is an end-to-end learning-based 3D point cloud registration framework that achieves comparable registration accuracy to prior state-of-the-art geometric methods of aligning two different point clouds.

Instead of implementing other keypoint based methods where a RANSAC procedure is usually needed, the implementation of various deep neural network structures is done to establish an end-to-end trainable network. The keypoint detector is trained through this end-to-end structure which enables the system to avoid the inference of dynamic objects and leverages the help of sufficiently salient features on stationary objects, thereby achieving high robustness.

Main Contributions

DeepVCP Visualization

Results demonstrate that it achieves comparable registration accuracy and runtime efficiency compared to state-of-the-art geometry-based methods, but with higher robustness to inaccurate initial poses.

Low registration error and high robustness of this method makes it suitable for substantial applications based on point cloud registration.