Image-based memory for robot navigation using properties of omnidirectional images

Abstract

This paper proposes a new technique for vision-based robot navigation. The basic framework is to localise the robot by comparing images taken at its current location with reference images stored in its memory. In this work, the only sensor mounted on the robot is an omnidirectional camera. The Fourier components of the omnidirectional image provide a signature for the views acquired by the robot and can be used to simplify the solution to the robot navigation problem. The proposed system can calculate the robot position with variable accuracy ("hierarchical localisation") saving computational time when the robot does not need a precise localisation (e.g. when it is travelling through a clear space). In addition, the system is able to self-organise its visual memory of the environment. The self-organisation of visual memory is essential to realise a fully autonomous robot that is able to navigate in an unexplored environment. Experimental evidence of the robustness of this system is given in unmodified office environments.

Local visual homing by matched-filter descent in image distances

“…With the three methods presented in what follows we intend to illustrate the broader applicability of the matched-filter concept. The first two methods adhere to the concept of DID but use a different distance measure and a different template flow field, respectively, whereas in the third example we derive the matched-filter solution for the Fourier-amplitude method (Menegatti et al 2004).…”

“…Fourier-amplitude method In the Fourier-amplitude method introduced by Menegatti et al (2004), each image is characterized by the first K amplitudes of the Fourier spectra of each row of the panoramic image. Since Fourier amplitudes are invariant against shifts, this method does not require a compass to align the two images before computing the distance.…”

“…In the ALV model, for example, f would produce the two components of the average landmark vector (Lambrinos et al 2000); in the contour model, it would amount to a computation of contour length and contour eccentricity (Möller 2001); in the Fourier-amplitude model, f would deliver the first few Fourier amplitudes for each image row (Menegatti et al 2004). …”

“…While usually in parameter methods the distance measure or "potential" can be computed but the home vector has to be estimated by sampling the potential at multiple locations in space, the ALV method directly provides a home vector (it is equal to the difference between the two average landmark vectors), whereas the potential cannot be computed because it depends on the unknown distances to the landmarks (Möller 2002). While these parameter methods are of mostly theoretical interest, the Fourier-amplitude method suggested by Menegatti et al (2004) may indeed be suitable for realworld applications. In this method, a panoramic image (with cyclic rows) is characterized by the first few Fourier amplitudes of each row.…”

“…Applying the matched-filter concept to the parameter method by Menegatti et al (2004) was a suggestion by Sebastian Fischer.…”