Dataset Open Access
Newey, Charles; Jones, Owain; Dee, Hannah
A simple computer vision dataset for shadow detection and texture analysis, specifically created to help test shadow detection algorithms (and texture segmentation algorithms) for mobile robots - that is, shadow detection with an active (moving) camera.
The dataset is focused around texture analysis, so each image sequence contains shadows moving in front of a number of various textured surfaces. The dataset contains four main subfolders: "active", "artificial", "kondo", and "static". The "static" folder contains ground-truthed image sequences of textured surfaces with shadows moving over them, and the "active" folder contains ground-truthed image sequences of a camera travelling over textured surfaces. The "artificial" folder contains a computer-generated 3D scene with computer-generated ground truth, but note that texture is absent from all images within. Finally, the "kondo" folder contains a series of extremely challenging images captured from a webcam mounted to a Kondo bipedal robot. This final dataset is challenging because it contains a high level of noise, flicker and interference from electrical lighting, and the poor lighting conditions make for complex shadows with large penumbrae.