The need for precise digital representations of complicated spatial environments has enabled 3D reconstruction to become increasingly vital in robotics, virtual reality, and even in the analysis of indoor spaces. While standard neural radiance fields (NeRFs) operate impressively in scene representation, their computational cost, memory deficits, and inflexibility to dynamic and changing environments limit their practicality. This paper addresses an advanced 3D reconstruction system that overrides the Fast NeRF renderer scene limitations using memory optimization techniques of Instant Neural Primitives and dynamic scene processing neural networks of D-NeRF. To evaluate the quality of the final rendered scene, peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and learned perceptual image patch similarity (LPIPS) are calculated. The changes to these metrics from training show how well the system has learned to capture structural and perceptual details and thus reinforce the system’s capability for high-quality dynamic 3D scene reconstruction. The implemented system is based on the PyTorch framework and optimized using CUDA for real-time rendering of dynamic scenes. Other systems for spatial analysis and visualization such as Open3D, Trimesh, and Pyrender are also used. InstantNGP has enabled significant memory savings.