A Literature Review On Semantic Segmentation Of 3D Temporal Data Using Neural Networks

Abstract

The fusion of 3D spatial information with temporal dynamics has given rise to a myriad of applications across diverse domains, from autonomous vehicles and medical imaging to robotics and beyond. The task of semantic segmentation, specifically applied to 3D temporal data, plays a pivotal role in deciphering intricate volumetric time series datasets. In this review, we provide a comprehensive overview of the state-of-the-art methodologies leveraging neural networks for semantic segmentation in the dynamic context of 3D temporal data. We begin by establishing the foundational principles of semantic segmentation and the unique challenges posed by the temporal dimension. It systematically reviews the evolution of neural network architectures, encompassing recurrent neural networks (RNNs), long short-term memory networks (LSTMs), 3D convolutional neural networks (3D CNNs), and emerging hybrid models. The exploration extends to attention mechanisms, spatiotemporal networks, and novel architectures that address the specific nuances of 3D temporal data. Drawing upon an extensive array of references spanning autonomous driving, medical image analysis, robotic navigation, and beyond, we categorized and evaluated these methodologies. The critical analysis encompasses factors such as accuracy, efficiency, scalability, and adaptability to various application domains. The insights provided in our review serve as a valuable resource for researchers, practitioners, and enthusiasts aiming to navigate the complexities of semantic segmentation in volumetric time-series datasets. By synthesizing theoretical foundations with practical applications, the survey aims to foster a deeper understanding of current state-of-the-art and inspire innovative approaches for future research.