3D Modeling and Rendering with a Tesla Model 3 Highland

Abstract. This paper investigates the feasibility of using video sequences recorded by a Tesla Model 3 Highland for photogrammetric 3D reconstruction and neural rendering. The onboard cameras, originally designed for autonomous navigation, were calibrated as a multi-camera rig using bundle adjustment. The resulting intrinsic and extrinsic parameters were validated across several test projects and subsequently applied to real-world driving sequences to generate oriented image datasets, 3D mesh reconstructions, and gaussian splatting renderings. The experiments demonstrate that complex scenes can be reconstructed, although artefacts persist due to limited acquisition geometry, temporal desynchronization, compression, and dynamic scene elements. The study highlights the photogrammetric potential of consumer vehicles and provides a quantitative evaluation of Tesla Vision data for 3D applications, addressing limitations, achievable accuracy, and prospects for automated artefact correction and large-scale reconstruction from vehicular fleets. A video with selected examples is available at https://youtu.be/pOFpUp-vlHU.