Human vision relies on mechanisms that respond to luminance edges. As a result, visual percepts fade if we spatially or temporally interfere with edge-sensitive processes in the visual system. Standard vision models focus on spatial mechanisms for human edge detection assuming that the modeled processes occur within a single fixation and under static viewing. Inspired by cells in V1, these multiscale vision models are comprised of linearly operating filters at multiple scales and with different orientations. However, the assumption of static viewing is not doing justice to the fact that our eyes are constantly moving and that the visual system is fundamentally driven by spatiotemporal information. Here, we propose a spatiotemporal model of human edge detection which combines multiscale vision with an active sampling strategy via fixational eye movements. In simulations, we show that in our model edge signals naturally emerge as a byproduct of actively sampling the visual input via ocular drift. To test our model, we conducted two simulation experiments: In the first experiment, we propose that our model can indirectly account for the spatial-frequency-specific effect of narrowband noise in human lightness perception. Second, we show that our model can extract edge signals in natural images and compare its performance to several controls. We argue that in a system that performs multiscale spatial filtering, extracting edge signals via the proposed spatiotemporal mechanism could reduce the number of oriented filters needed in V1 that process the input with both different orientations and spatial frequencies. [Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy – EXC 2002/1 "Science of Intelligence" – project number 390523135.]