Efficient Real-Time Sign Detection for Autonomous Vehical in Hazy Environment Using Deep Learning Models

The reliability of traffic sign detection and recognition in (TSDR) is critical to the effective implementation of autonomous vehicle technology. While several approaches for TSDR have been presented, most have been assessed on clean datasets, ignoring the degradation performance associated with extreme climates (CCs) that hide traffic sign photos collected in the outdoors. We provide a system using deep learning for effective real-time traffic sign identification in adverse weather situations in this study. Our method comprises dark channel hazing and picture conversion and dehazing with YOLOv3 and YOLOv5. We present a modular solution based on Convolutional neural network models (CNN), which includes a challenge classifier, an encoder-decoder CNN for picture enhancement, and different CNN architectures for sign recognition and classification. To boost detection accuracy, we concentrate on improving the traffic sign areas in tough photos. Our technique is tested using the GTSRB dataset, which features traffic recordings collected under various CCs. Using the German Traffic Sign Recognition Benchmark (GTSRB) dataset, the reported technique attained an accuracy of 98.62 In addition, we compare our method to various CNN-based TSDR approaches, indicating its superiority.