Defocus data for automotive driving scenario

The dataset contains images captured with a tripod-mounted automotive camera set to approximate a car’s front-camera height and aligned down the center of a lane toward a vehicle ahead. It depicts a static street-crossing scene with three object classes: person (a mannequin for consistency), bicycle, and car. Two cars may appear: the first car is closer to the camera and fully visible when in view; the second car is farther ahead near the VRU and may be partially occluded by the first car at longer distances. At distances ≤7 m, only the second car is in the camera’s field of view. The VRU is stationary and captured at four positions relative to the second car: far-side flank, middle, near-side flank, and 1 m beyond the car, producing varying levels of occlusion from barely visible to fully visible. Camera-to-VRU distance ranges from 2–100 m, grouped as near (2, 3, 5, 7 m), mid (10, 15, 25 m), and far (40, 60, 80, 100 m). For each distance there are two scenes, each with four VRU positions, and each position includes nine defocus levels (5 negative, 3 positive, 1 zero/in-focus). Objects remain stationary while all defocus images at a position are captured; only placements change between captures. Scene 1 has only pedestrian (dummy) and scene 2 has pedestrian with a bicycle.Each image is named in the following pattern: <Date>T<Timestamp>_ Z<scene number>_<distance to the VRU(dummy)>_<position number>_<defocus position in micrometers>_coarse integration time (coarse_it)_<value of coarse_it>_fine integration time (fine_it)_<value of fine_it>.pngA curated version of this dataset was used in the paper (more details about the camera setup and dataset can be found in the paper): Estimating the effect of age-induced defocus on object detection performance of automotive cameras