The objective of creating this dataset to train AI models for traffic analysis or autonomous driving. This dataset contains 150 images which were taken from YouTube. This video showed the busy streets of Chawk Bazar, Dhaka, Bangladesh during daytime. The images were annotated with Computer Vision Annotation Tool (CVAT) with labels such as rickshaw, car, motorcycle, pedestrian, leguna, truck, and bus.
The dataset is available in Kaggle: BD traffic dataset
The link of the video is given below:
4K Bangladeshi Walking Tour Dhaka City 2023 || Dhaka, Chawk Bazar 4K Walking Tour 2023
The video shows the busy streets of Chawk Bazar, Dhaka, Bangladesh, which are suitable for building traffic control or autonomous driving systems. The resolution of the video is 4K. 150 high-quality images were extracted. The images were extracted in 1080p resolution.
I used CVAT for annotation. The reason is that this tool is full of useful features that can handle annotations efficiently. After annotation, we can export the annotated dataset to any format compatible with any AI model.
I chose eight labels for annotation. The reason behind choosing these labels is because they are a common scenario in the streets of Bangladesh. The labels are given below:
- Rickshaw
- Car
- Motorcycle
- Pedestrian
- Leguna
- Truck
- Bus
- CNG
Bounding boxes were drawn around each object in every image. Tight bounding boxes were used around the object so that the AI model can easily understand the object. Finally, annotated images were exported in Pascal VOC format.
The objects frequently overlapped with each other. So, I have to annotate carefully to separate the objects correctly. Secondly, I tried to use the Python library to download the YouTube video. The plan was to convert the video into frames and extract 150 images. But the plan didn’t work because I couldn’t download the video higher than 360p. As a result, I took screenshots from the video and annotated them.