Abstract
Due to a slow acquisition, the spatial resolution of a depth-acquisition device, such as a Light Detection and Ranging (LiDAR) sensor, is limited, which strictly requires scanning a field of view (FOV) in a particular order. To accurately reconstruct a depth image from limited spatial samples, two-stage sampling has been widely used. However, from the perspective of a LiDAR, two-stage sampling requires scanning the FOV two times to build a depth map, and eventually becomes counterintuitive in practice. To address this problem, this study presents a LiDAR framework and a gradient-aware line-based sampling algorithm. Different from previous works, the proposed scanning algorithm allows a LiDAR to scan the FOV from top to bottom while simultaneously seeking sampling points along object boundary. By utilizing the information of the previous line during scanning, the proposed method maintains the conventional scanning order in a LiDAR, while efficiently predicting the sampling locations of the current line. The experimental results demonstrate that the proposed sampling outperforms grid sampling by at most 9.79 dB on the synthetic datasets. Consequently, the proposed sampling achieves reconstructed quality similar to that of optimal sampling in the previous design, while substantially reducing the computation time and memory requirements. The experimental results with the laser range data and the real data captured by the LiDAR system demonstrate that the proposed method can reduce the averaged mean-absolute-error (MAE) by 34.91%, 47.23%, 54.88%, and 57.99% for the sampling ratios of 20%, 30%, 40%, and 50%, respectively, compared to the conventional LiDAR sampling.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have