BanglaRoad: A Multi-Environment Road Scene Dataset from Bangladesh for Object Detection in Autonomous Navigation Systems
Description
Description: This dataset, named BanglaRoad, introduces a labeled road scene dataset collected from three distinct road environments in Bangladesh: rural roads in Sirajganj district, urban roads in Dhaka, and highway corridors along the Sirajganj-Dhaka route. It contains 1,440 images (post-augmentation) annotated with bounding boxes across 15 object classes. The dataset includes Bangladesh-specific categories such as rickshaw, CNG auto-rickshaw, Nosimon (a locally fabricated agricultural vehicle), pothole, and free-roaming cattle, which are absent in standard international benchmarks. The dataset is intended to improve object detection models for autonomous driving in unstructured and complex traffic environments typical of South Asia.
Files
Steps to reproduce
Road images were collected using mobile cameras mounted in vehicles and held by pedestrians during 2025-2026. The raw dataset was organized into three categories: rural roads (Sirajganj Sadar, Shahzadpur, Belkuchi), urban roads (Dhaka), and highways (Sirajganj-2 Lane). All annotations were created using Roboflow's web-based platform, drawing bounding boxes tightly around the visible extent of each object across 15 distinct classes. Images were preprocessed through Roboflow using Auto-Orient to correct EXIF rotation and Resized to 640x640 pixels. Offline augmentation was applied via Roboflow to expand the training set.
Institutions
- Khwaja Yunus Ali UniversityRajshahi Division, Chauhāli