EventSTR introduces the first large-scale benchmark dataset for event-based scene text recognition, leveraging bio-inspired cameras to overcome traditional limitations.

• Works effectively in challenging conditions including low light and motion blur
• Contains 9,928 high-definition event samples with both Chinese and English text
• Establishes baseline methods for this emerging recognition approach
• Enhances security applications by enabling text recognition where traditional cameras fail

This research opens new possibilities for security systems that require reliable text recognition in surveillance scenarios with poor visibility or dynamic environments.

EventSTR: A Benchmark Dataset and Baselines for Event Stream based Scene Text Recognition