PDC: Piecewise Depth Completion utilizing Superpixels
This addresses depth completion for autonomous driving systems by offering an alternative to CNN-based methods that avoids issues like flying pixels and overfitting.
The paper tackles depth completion from sparse LiDAR and RGB data for autonomous driving by proposing PDC, a non-deep learning method that segments images into superpixels and groups them using a cost map, achieving state-of-the-art accuracy on the KITTI dataset.
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.