Improving Radar-Camera Fusion Network for Distance Estimation

In today's world, a sensor-based autonomous industry needs a range of sensors to understand its environment and decide how to function safely and productively. The precise and trustworthy assessment of the surroundings around a vehicle has made a variety of autonomous or assisted driving approaches possible. Hence, we provide a novel radar camera fusion technique based on deep learning that greatly enhances object identification performance. The combination of radar and camera data in particular has emerged as a promising strategy for improving the precision and dependability of object identification systems. However, the performance of current fusion algorithms is frequently constrained by issues with occlusion, variable illumination, and sensor noise. With enhancing capabilities of autonomous vehicles through different algorithms and perception methodologies, it has become important to look into the importance of computationally lighter algorithms which work efficiently as well. Based on the knowledge gleaned from analyzing several deep learning models, we came up with a modified fusion algorithm by adding a number of new operations. The processes of this algorithm consist of improved object identification techniques, feature extraction, and feature fusion. Our approach successfully addresses the issues posed by occlusion, varying illumination, and sensor noise by utilizing the complementary nature of radar and camera technologies, resulting in enhanced object detection performance. In conclusion, this study offers a cutting-edge deep learning-based radar camera fusion technique that surpasses current methods for object detection. Our approach provides improved accuracy, robustness, and real-time performance by using insights from several deep-learning models and including extra processes. The suggested approach provides a sizable improvement in sensor fusion methods, opening the door for more improvements to perception system enhancement.