Indian Heterogeneous Traffic Dataset with 5 Vehicle Classes from Kerala

A dataset of annotated traffic keyframes for vehicle detection, featuring long-tailed class distributions and multi-vehicle occlusions. It contains 5 distinct vehicle classes, including regional modes like autorickshaws, with data captured from surveillance systems at junctions in Kozhikode District, Kerala, India. The dataset was manually annotated using CVAT at the National Institute of Technology Calicut and is structured in YOLO format for deep learning integration.

Use Cases

• Training vehicle detection models based on annotated bounding boxes for 5 vehicle classes.
• Developing models robust to severe occlusions based on the described multi-vehicle occlusion challenges.
• Analyzing traffic flow and composition in Indian urban environments based on the regional vehicle classes like autorickshaws.
• Benchmarking object detection algorithms on datasets with long-tailed class distributions as mentioned in the description.

Strengths

• Includes 5 distinct vehicle classes, specifically highlighting regional transportation modes like three-wheeled autorickshaws.
• Data was manually bounded and verified for high precision using the CVAT annotation tool.
• Dataset is structured in the standard YOLO Object Detection format for seamless integration with frameworks like YOLOv8.
• Real-world keyframes were captured using high-angled urban surveillance and traffic monitoring ANPR systems.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.
• Data is geographically focused on Kozhikode District, Kerala, which may limit generalizability to other regions.

Provenance

Source

Raw keyframes captured from traffic surveillance systems deployed across primary junctions in Kozhikode District, Kerala, India, with permission from Kozhikode District Police.

Collection Method

Data instances were manually bounded and verified using the CVAT (Computer Vision Annotation Tool) platform.

Freshness

Last updated 2026-05-24 11:29:12; freshness should be verified.

Geography

Kozhikode District, Kerala, India