How We Built a Real-World Bangladesh Rickshaw Traffic Dataset for Physical AI

#ai #computervision #machinelearning #opensource

Open datasets are essential for advancing AI, but truly diverse real-world traffic data remains difficult to find.

To help address this gap, we built a Bangladesh Rickshaw Traffic Dataset designed for Physical AI, Computer Vision, Robotics, and Autonomous Driving research.

The dataset captures complex urban traffic scenes where rickshaws, motorcycles, buses, trucks, pedestrians, and private vehicles interact naturally in challenging environments.

Why Bangladesh?

Most publicly available traffic datasets are collected in structured road environments where traffic behavior is relatively predictable.

Bangladesh presents a very different scenario.

Urban roads are shared by rickshaws, motorcycles, buses, trucks, private vehicles, street vendors, bicycles, and pedestrians. These complex interactions make the dataset particularly valuable for training AI systems that must operate in diverse real-world environments rather than controlled conditions.

This diversity helps improve model robustness and generalization across different traffic scenarios.

Dataset Highlights

Some key characteristics of the dataset include:

Real-world urban traffic captured in Dhaka
Diverse traffic participants including rickshaws
Human Ground Truth annotations
Rich metadata for AI research
Privacy-protected preview samples
Commercial-ready organization
Quality validation pipeline
Dataset documentation

Metadata Design

We believe metadata is just as important as the video itself.

Each clip is accompanied by structured metadata that helps researchers filter, search, and organize data efficiently.

Examples include:

Scene category
Road context
Object statistics
Annotation status
Quality indicators
Privacy processing
Capture information
Dataset organization

This reduces preprocessing time and makes the dataset easier to integrate into machine learning workflows.

Why We Opened a Free Sample

Many researchers and companies hesitate to evaluate datasets before purchasing them.

To make evaluation easier, we released a free sample through multiple platforms so users can inspect the data quality before requesting larger commercial datasets.

Our goal is transparency and long-term collaboration with the AI community.