Wheelchair Users Dataset

dsContentTTds

Content


Tumenthis dataset was recorded by the Video Processing and Understanding Lab due to the lack of public wheelchair datasets. We used it to test the trained wheelchair users detector, as it contains sequences with a higher number of wheelchairs (up to four) and some more complex situations and scenarios (illumination changes, occlusions, etc.). The sequences were recorded in a real environment of a senior residence, in order to work with an environment as realistic as possible (due to privacy issues, real recording with actual residents was not possible). Each of the frames has a resolution of 768x432 pixels and the sequences are recorded at 25 fps. Compared to the other dataset, this one contains a new environment with a larger number of sequences, a greater number of frames per sequence, and more wheelchair types (three different wheelchairs). The dataset consists of 11 sequences (S1 to S11), each of them recorded from two points of views (V1 and V2), resulting in a total of 22 sequences.

All sequences were recorded in the same room, using two GoPro cameras (HERO3 White edition). The fisheye effect was corrected using the GoPro Studio software tool.