Vision-Based Vehicle Localization Using a Visual Street Map with Embedded SURF Scale

Abstract

Accurate vehicle positioning is important not only for in-car navigation systems but is also a requirement for emerging autonomous driving methods. Consumer level GPS are inaccurate in a number of driving environments such as in tunnels or areas where tall buildings cause satellite shadowing. Current vision-based methods typically rely on the integration of multiple sensors or fundamental matrix calculation which can be unstable when the baseline is small. In this paper we present a novel visual localization method which uses a visual street map and extracted SURF image features. By monitoring the difference in scale of features matched between input images and the visual street map within a Dynamic Time Warping framework, stable localization in the direction of motion is achieved without calculation of the fundamental or essential matrices. We present the system performance in real traffic environments. By comparing localization results with a high accuracy GPS ground truth, we demonstrate that accurate vehicle positioning is achieved.

Position Interpolation Using Feature Point Scale for Decimeter Visual Localization

“…This process makes use of feature scale, so is most easily applicable to methods which already use scale-invariant features [11], [23] (and also a variant of the method presented by Badino et al [2] which uses features instead of WISURF, as introduced in the same paper). In particular it is applicable to systems using feature scale [23]. The size, or scale, of extracted feature points is related to the distance between the capture location and the real-world position of the feature.…”

“…This system does however still require calculation of the essential matrix. Alternatively, the scale change between matched features can be compared and used as a similarity measure [23]. Feature points that have a size or scale property, such as the Scale Invariant Feature Transform (SIFT) [14], will vary in size depending on the distance from which they are captured by the camera.…”

“…As the query image capture location becomes close to the corresponding database image, the scale of matched features will also become closer. The scale difference can be used as a cost measure in DTW [23].…”

“…In addition, the feature matching process of comparing many feature descriptors is a computationally intensive process. In order to overcome these performance bottlenecks, feature-based localization methods which avoid calculation of image geometry [2], [23] use either descriptor distance [2] or feature scale [23] to determine the closest database image directly.…”

“…Localization can either be performed using extracted feature points [11], [12], [23], or by matching query images to database images using whole image similarity [2], [6], [22], [24]. The correct database image match can be determined either using Dynamic Time Warping (DTW) with an image similarity measure [11], [22], [23] or a Bayes filter [2], [6], [24]. Methods using feature points have the advantage of robustness to occlusions and lane changes, but also come with increased complexity.…”