Autonomous driving paper index
Physically-informed neural fields for non-line-of-sight transient reconstruction
One-line summary
In this paper, an improved NLOS imaging reconstruction algorithm based on neural transient fields is proposed.
Engineering notes
Key topics: autonomous driving. See the paper for implementation details and experimental results.
Chinese explanation / 中文解读
中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。
Original abstract
Non-line-of-sight (NLOS) imaging recovers occluded targets beyond line of sight, with important applications in autonomous driving, reconnaissance, and emergency rescue. Conventional NLOS imaging algorithms exhibit limitations including slow convergence, insufficient recovery of high-frequency details, and high sensitivity to noise artifacts under sparse sampling. In this paper, an improved NLOS imaging reconstruction algorithm based on neural transient fields is proposed. Specifically, positional encoding in the network is replaced with hash encoding to enhance training efficiency and high-frequency detail representation. Gaussian noise is introduced to simulate realistic detection errors and implement model regularization, so that noise and artifacts under sparse sampling can be suppressed. A two-stage training strategy and hierarchical sampling scheme are integrated to further improve reconstruction accuracy and robustness. The proposed method effectively addresses reconstruction challenges in low signal-to-noise-ratio and complex occlusion scenarios, providing a feasible solution for the practical application of NLOS imaging technology.
Links and sources
Need this topic turned into a technical roadmap?
Full Self Driving can prepare a custom autonomous driving literature review, code map, dataset map, and B2B technology assessment.
Request B2B research
Comments