A Sliding Window Technique-Based Radar and Camera Fusion Model for Object Detection in Adverse Weather Condition

The limits of individual sensors have led to a recent trend of preferring sensor fusion for increased object detection and tracking in Advanced Driver Assistance Systems (ADAS), which mainly rely on sensors. Although radars are pretty good at measuring depth and speed, more is needed to identify objects precisely. On the other hand, cameras have trouble seeing depth and operate poorly in low light, rain, fog, and dust. To overcome these challenges, this paper introduces a novel fusion technique that fuses radar and camera data for enhanced object detection and classification using the sliding window method. This technique identifies and validates regions of interest within the merged radar and camera imagery. To handle the fused inputs efficiently, the proposed framework integrates layers to a deep neural network (DNN) architecture that was influenced by GoogLeNet's fully convolutional networks (FCN). This method extracts critical features by applying the sliding window mechanism on the input images. The evaluation framework integrates Time Division Multiplexing (TDM) and Frequency Modulated Continuous Wave (FMCW) radar with a monocular camera. Tests were conducted under various weather conditions to demonstrate the model's utility in Automated Emergency Braking (AEB) scenarios through behavioural analysis. The results show that the model is very suitable for real-time applications in autonomous vehicle systems since it is not only computationally efficient but also responds quickly..