Detection and classification of road signs in raining condition with limited dataset

The road sign recognition (RSR) system is used to complete two tasks: localizing the traffic in an image and then classifying it according features. Some of applications such are incorporated into Advanced Driver Assistance System (ADAS) autonomous vehicles. However, accuracy model decreases when changes lighting or weather occurs, lack training samples taken rainy condition causes be sub-optimal. research was conducted with focus on solving problems, object detection data, especially images under adverse condition. In this work, we compare analyze three methods; automatic white balance (AWB), policy augmentations Image-to-Image-translation (I2IT) technique their performance detect signs raining conditions. All methods were built upon pre-trained SSD-MobileNetV2 using TensorFlow2 framework. from Malaysia Traffic Sign Dataset (MTSD) train all models. Finally based result a final combination proposed that achieved best Experimental results showed AWB not effective detecting condition, while other techniques highly effective. implemented by combining augmentation I2IT, obtained mAP 0.7967 clear 0.7160 added testing dataset. These corresponds at 50% IoU [email protected] during [email protected] images, which outperformed Thus, classification can perform well limited dataset has been successfully developed.