A review of multisensor data fusion techniques for reliable autonomous perception

Autonomous systems such as self-driving vehicles, drones, and intelligent robots rely on accurate perception to interact safely with dynamic environments. However, single-sensor systems are prone to noise, occlusion, and environmental variability, limiting their reliability. Multisensor Data Fusion (MSDF) has therefore emerged as a crucial approach for improving perception by integrating complementary information from multiple sensors, including cameras, LiDAR, radar, GPS, and inertial measurement units (IMUs). This paper reviews current MSDF techniques and their role in achieving reliable, robust, and real-time autonomous perception. The review categorizes fusion approaches into three major groups: classical probabilistic models, knowledge-based reasoning techniques, and data-driven deep learning methods. Classical models such as Kalman Filters, Extended Kalman Filters, Particle Filters, and Bayesian networks provide strong statistical foundations for uncertainty estimation and sensor integration. Knowledge-based systems, including Dempster–Shafer theory and fuzzy logic, effectively manage imprecise or conflicting sensory information. More recently, deep learning-based fusion techniques, leveraging Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and attention-based models, have enabled end-to-end learning from multimodal inputs, achieving higher perception accuracy and adaptability in complex environments. A comparative analysis reveals that classical fusion methods offer transparency and low computational cost but are limited in scalability and nonlinearity handling. Deep learning approaches, while powerful, face challenges related to explainability, computational demand, and data dependency. Finally, the study highlights emerging trends shaping the future of autonomous perception, including edge AI and distributed fusion, explainable AI (XAI) for decision transparency, and the integration of 5G/6G communication for cooperative sensing.