PDC: Piecewise Depth Completion utilizing Superpixels

TL;DR

The paper introduces PDC, a non-deep-learning method for depth completion that segments images into superpixels and uses a cost map to improve depth accuracy, demonstrating state-of-the-art results on KITTI.

Contribution

PDC is a novel piecewise depth completion approach that avoids deep learning, addressing issues like depth discontinuities and overfitting present in CNN-based methods.

Findings

Achieves state-of-the-art accuracy on KITTI dataset.

Effectively handles depth discontinuities with superpixel segmentation.

Demonstrates the effectiveness of non-deep-learning depth completion.

Abstract

Depth completion from sparse LiDAR and high-resolution RGB data is one of the foundations for autonomous driving techniques. Current approaches often rely on CNN-based methods with several known drawbacks: flying pixel at depth discontinuities, overfitting to both a given data set as well as error metric, and many more. Thus, we propose our novel Piecewise Depth Completion (PDC), which works completely without deep learning. PDC segments the RGB image into superpixels corresponding the regions with similar depth value. Superpixels corresponding to same objects are gathered using a cost map. At the end, we receive detailed depth images with state of the art accuracy. In our evaluation, we can show both the influence of the individual proposed processing steps and the overall performance of our method on the challenging KITTI dataset.