WalkitDatabase

In this work a multi-camera, multi-modal and detailed dataset for vision-based applications was constructed and its process extensively detailed. Using a smart walker designed for gait rehabilitation therapy sessions, equipped with a pair of RGB+D cameras, color and depth data was acquired from a total of 14 healthy subjects, at 3 different walking velocities and in 3 different walking scenarios/paths, repeated 3 times in different locations. Simultaneously, accurate skeleton data was recorded and synchronized using an inertial-based commercial motion capture system that provides a reliable ground-truth for classical or novel (i.e. machine learning-based) vision applications. A procedure to correlate each color and depth frame with the external ground truth skeleton joint data is also described. In total, the database contains ~370k frames of synchronized data sampled at 30Hz, which amounts to ~205 minutes of total recording time.