Abstract:
Traffic monitoring, driver assistance, and autonomous driving systems all depend ontheidentification of road signs. Road sign detection is vital for safety and advanced systems
for drivers. In this context, this study is aimed at establishing a development of a deeplearning approach for Bangla Road sign recognition using CNN. The dataset includes
images linked to the 11 categories of Bangla Road signs such as “Traffic Merges fromLeft,” “Speed Limit 40 km,” among others. In this study, we compare the performance of
VGG16, Xception, DenseNet201, ResNet50 and a CNN model developed particularlyfor
this purpose to accurately identify these signs. The potential CNN architecture that the wepropose is tailored for the issues of Bangla script and the enhancement of text-based roadsign recognition. Frameworks such as TensorFlow and Keras were employed andthetraining and testing were performed on GPUs for faster computation. Conclusively, theCNN sees an accuracy of 96.41% by subjecting it to a process of testing/evaluationthat
considers it comparable with the benchmark models including VGG16 (99. 51%), Xception (99. 51%), DenseNet201 (96. 97%), and ResNet50 (96. 083). This studyis
useful in real life such as affairs related to traffic control and Self-driving cars in terms of
CNN in Bangla Language. It adds knowledge to techniques in model enhancement, dataset acquisition, and performance assessment methodologies, which will open the door
for further research on intelligent transportation systems based on deep learningtechnology.
