Implicit Scene Representation Reconstruction Method Based on Jetson Nano
With the ongoing advancements in computer vision and edge computing technology, there is a growing focus on the efficient reconstruction and representation of complex scenes. This study introduces a method for implicit scene representation reconstruction using Jetson Nano, aiming to seamlessly integrate deep learning with edge computing technology. Leveraging the Jetson Nano edge computing platform and connecting an RGBD camera to it, the collected data is transmitted to the cloud. The study employs an innovative implicit representation model to compactly and efficiently represent scene information through functions. Optimized with Compute Unified Device Architecture ( CUDA) and Turing Tensor R-Engine( TensorRT) on the Jetson Nano platform, the method enhances computational efficiency. By combining implicit neural networks with simultaneous localization and mapping (SLAM) techniques, the research successfully achieves accurate 3D scene reconstruction, with an average root mean square error of absolute trajectory error in camera tracking reaching 1. 87 . The performance proves robust across various scenes.
万方数据
维普
