Vision-Based Aircraft Detection with 55,521 Images for Sense and Avoid

55,521 high-resolution images across 15 sequences depict a fixed-wing aircraft on a collision course with a stationary camera. The dataset, created by Martin, Riseley, and Ford in 2022, includes ground truth labels and video sequences. It is hosted on Hugging Face and was last updated on July 14, 2025.

Use Cases

• Training object detection models based on the high-resolution image sequences of approaching aircraft.
• Developing and testing collision prediction algorithms based on the aircraft's collision course trajectory.
• Benchmarking vision-based sense-and-avoid systems using the provided ground truth labels.
• Analyzing temporal patterns in aircraft approach using the sequential image and video data.

Strengths

• Contains 55,521 high-resolution images, providing substantial visual data.
• Includes 15 distinct image sequences, offering multiple scenarios for analysis.
• Provides ground truth labels and video sequences for comprehensive validation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset appears to focus on a single, specific scenario (stationary camera, fixed-wing aircraft), which may limit generalizability.

Provenance

Source

jasminmartin on Hugging Face, citing Martin, J., Riseley, J., & Ford, J. J. (2022).

Collection Method

Likely captured via a stationary, grounded camera filming an approaching aircraft.

Time Range

null

Freshness

Last updated 2025-07-14 01:20:50; freshness should be verified.

Geography

null