Depth Completion Using Classical Image Processing

This video showcases the results of our depth completion algorithm running at 90 Hz on the CPU (Gaussian Blur version). The algorithm consists of a set of consecutive image processing operations applied to fill in missing depth values in the projection of a LIDAR point cloud to image space. The output depth map of the 87 Hz Bilateral Blur version preserves local structure, and can benefit 3D object detection or SLAM algorithms that use point cloud input. For more information, please see our paper https://arxiv.org/abs/1802.00036