Real-time Depth Completion using Radar and Camera

One of the active challenges in intelligent driving systems is depth completion. The KITTI benchmark depth completion challenge is concerned with lidar-camera fusion. Recently, researchers attempted to address the problem using radar and camera fusion due to the common use of radar sensors in the automotive industry. In this study, we take a closer look at one of the developed models for radar-camera fusion. We propose introducing preprocessing algorithms for radar data such as CPU-based methods to improve the radar measurements and subsequently the depth estimation. The methods include dilation, closure, and hole filling. With the usage of late fusion between radar and camera, the quantitative results show that the proposed methods improve the estimation with almost no additional computational cost. Additionally, the qualitative analysis demonstrates how the proposed model performs better.