The NightOwls Dataset focuses on pedestrian detection at night. In the following, we give an overview on the design choices that were made to fit the purpose of the dataset and highlight the most important characteristics. You can find more details in our paper.
Features
Complexity
- 4 classes (distrbution)
- Pedestrians
- Bicycledriver
- Motorbikedriver
- Ignore areas
Diversity
- 3 countries and multiple European cities are represented
- All 4 seasons
- Dawn and nighttime
- All weather conditions
- Large number of dynamic objects
- Varying scene layout
- Varying background
- Quality realistically depends on scene illumination and the vehicle speed, including blurred and sharp images
Volume
- 279k frames out of 40 sequences (examples)
- All fully annotated or only background
- Additional information
- Pose
- Difficulty
- Occlusion
- Truncation
Challenges
- Motion Blur and image noise
- Reflections and high dynamics
- Large variation in contrast, reduced color information
- Weather and seasons
Metadata
- Industry-standard camera
- Image resolution 1024 x 640
- Frame-rate of 5,33fps by taking every third frame with an original frame-rate of 16fps