With the increasing power of technology, smart traffic control systems have become a vital research domain for every country. For any Intelligent Transport System (ITS), vehicle detection, localization, and classification are the main tasks. Large-scale databases are a key component of data-driven, sophisticated object detection models. The use of well-labeled datasets has considerably enhanced the efficacy and accuracy of many different object identification tasks. Building up a dataset of vehicles is hence the initial stage in detecting vehicles on Bangladeshi roads. Considering the scarcity of datasets for local vehicles in Bangladesh, we have proposed a new dataset called the “Vehicle BD” dataset for vehicle localization and categorization. This dataset is comprised of a total of 12,413 images. We have considered nine types of local vehicles: bus, truck, car, CNG, easy bike, motorcycle, bicycle, rickshaw, and van. Data labeling has been done by using “LabelImg”. For privacy purposes, we have also blurred the faces in the images. This dataset can be a benchmark for research in the context of the Bangladesh road scenario.