Computer Vision

End-to-end Sparse Correspondence Prediction for Relative Camera Pose Estimation

We present an end-to-end learnable, differentiable method for pairwise relative pose registration of RGB-D frames. Our method is robust to big camera motions thanks to a self-supervised weighting of the predicted correspondences between the frames. Given a pair of frames, our method estimates matches of points and their visibility score. A self-supervised model predicts a confidence weight for visible matches. Finally, visible matches and their weight are fed into a differentiable weighted Procrustes aligner which estimates the rigid transformation between the input frames.

Kinect Fusion: Dense Surface Mapping and Tracking

Implementation of the paper “Kinect Fusion: Real-Time Dense Surface Mapping and Tracking”

SLAM for Autonomous Vehicles

In this project, I worked on the SLAM pipeline for an autonomous driving vehicle.

The tools I used for this project are, ROS, C++, PCL library, Ceres Solver, and Google Cartographer.

In addition, I created a ROS package that generates a cost map from a point cloud generated by SLAM. A cost map is a matrix where each cell contains a cost value. This value expresses how likely it is that the cell is occupied by an object. The cost value is also used to express “preferred” surfaces: for example, asphalt is preferred over grass.