Biologically plausible visual homing methods based on optical flow techniques

Abstract

Insects are able to return to important places in their environment by storing an image of the surroundings while at the goal, and later computing a home direction from a matching between this 'snapshot' image and the currently perceived image. Very similar ideas are pursued for the visual navigation of mobile robots. A wide range of different solutions for the matching between the two images have been suggested. This paper explores the application of optical flow techniques for visual homing. The performance of five different flow techniques and a reference method is analysed based on image collections from three different indoor environments. We show that block matching, two simple variants of block matching and two even simpler differential techniques produce robust homing behaviour, despite the simplicity of the matched features. Our analysis reveals that visual homing can succeed even in the presence of many incorrect feature correspondences, and that low-frequency features are sufficient for homing. In particular, the successful application of differential methods opens new vistas on the visual homing problem, both as plausible and parsimonious models of visual insect navigation, and as a starting point for novel robot navigation methods.

Local visual homing by matched-filter descent in image distances

“…Visual homing methods received attention from neuroethology as models for the navigation abilities of social insects and from robotics as computationally cheap building blocks for topological approaches to map building and navigation (Franz et al 1998a). Since these methods have been extensively reviewed in recent publications (Vardy and Möller 2005;Zeil et al 2003;Franz and Mallot 2000), we restrict ourselves to the brief classification shown in Fig. 1.…”

“…In block matching, the matching is based on the pixel-by-pixel distance between two rectangular regions in two images, in intensity matching, the block is reduced to a single pixel, and gradient matching searches for the best-matching gradient (Vardy and Möller 2005). Differential flow methods are derived from Taylor approximations of correspondence equations between intensities (first-order) or between gradients (second-order) under the assumption of small shifts (Barron et al 1994).…”

“…Differential flow methods are derived from Taylor approximations of correspondence equations between intensities (first-order) or between gradients (second-order) under the assumption of small shifts (Barron et al 1994). Vardy and Möller (2005) reported that flow-based matching methods and differential flow methods yield surprisingly robust navigation behavior. This is especially surprising for differential flow methods, since the above-mentioned assumption of small shifts is violated for the long spatial distance between the target and the current location.…”

“…Also in the matching methods the search was restricted to the vicinity of a region; thus some regions will have moved beyond the search range. Vardy and Möller (2005) explain this good performance by the fact that, even though the flow fields are of relatively low quality, there is still a sufficiently large number of flow vectors, particularly in the vicinity of the foci of expansion and contraction, that fulfill the assumption of small shifts and therefore produce a concordant vote for the direction of the home vector, whereas the erroneous flow vectors yield uncorrelated movement directions and are therefore overruled.…”

“…5.4). Despite these restrictions, the warping method produces very reliable homing behavior in different environments; it is only outperformed by some flow-based matching methods and differential flow methods that, however, require a compass to align the two images (Vardy and Möller 2005).…”