Fig 5. Building the space-time density volume from distance kernels around trajectory.

At each voxel layer, the algorithm finds the distance between each voxel (shown as a light-grey cube on the left) and the trajectory (in blue, distance shown with a green arrow), then assigns the value obtained from the kernel function at this particular distance in the kernel space (represented by the 3D coordinate system on the right) to the respective voxel. We link the kernel size (given by the radius h–the bandwidth) to the theory of foveal and parafoveal attention. We further assume that if a person is looking at a certain point, the attention away from the point will decrease linearly with distance and will be cut off at distance h, where h marks the boundary for foveal or parafoveal attention. We therefore use a linear kernel function in the form of a cone in kernel space.