Title: Simulation-driven optimisation of neural radiance fields for 3D traffic scene reconstruction

Authors: Dan Shan; Yaohui Sun; Xiangdong Meng; Yan Xing

Addresses: School of Electrical and Control Engineering, Shenyang Jianzhu University, Shenyang, Liaoning – 110168, China ' School of Transportation and Geomatics Engineering, Shenyang Jianzhu University, Shenyang, Liaoning – 110168, China ' School of Electrical and Control Engineering, Shenyang Jianzhu University, Shenyang, Liaoning – 110168, China ' School of Transportation and Geomatics Engineering, Shenyang Jianzhu University, Shenyang, Liaoning – 110168, China

Abstract: Accurate three-dimensional (3D) scene reconstruction is crucial for applications such as autonomous driving and traffic management, yet existing methods frequently exhibit aliasing and blurring in complex multi-view scenes. To overcome these challenges, we introduce an innovative deep learning algorithm that incorporates feature meshes into multi-scale neural radiance fields. This approach improves both the quality and computational efficiency of 3D scene reconstruction, facilitating detailed views and scalability to larger scenes. By optimising the projection of 3D rays and minimising the error between observed and rendered colours, our method substantially reduces training time and enhances key image quality metrics. Experimental results demonstrate a 43% reduction in training time and improvements in peak signal-to-noise ratio (PSNR) (6.3%), structural similarity (SSIM) (4.5%), and image perceived similarity (LPIPS) (20%) compared to existing techniques, rendering it a practical solution for real-world traffic scene reconstruction.

Keywords: neural radiance fields; NeRFs; 3D traffic scene reconstruction; anti-aliasing; rendering optimisation; autonomous driving; deep learning; multi-scale rendering; performance metrics.

DOI: 10.1504/IJSPM.2024.146029

International Journal of Simulation and Process Modelling, 2024 Vol.21 No.4, pp.275 - 293

Received: 08 Nov 2024
Accepted: 13 Feb 2025
Published online: 01 May 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article