Image Based Geo-localization in the Alps

Abstract

Given a picture taken somewhere in the world, automatic geo-localization of such an image is an extremely useful task especially for historical and forensic sciences, documentation purposes, organization of the world's photographs and intelligence applications. While tremendous progress has been made over the last years in visual location recognition within a single city, localization in natural environments is much more difficult, since vegetation, illumination, seasonal changes make appearance-only approaches impractical. In this work, we target mountainous terrain and use digital elevation models to extract representations for fast visual database lookup. We propose an automated approach for very large scale visual localization that can efficiently exploit visual information (contours) and geometric constraints (consistent orientation) at the same time. We validate the system at the scale of Switzerland (40,000 km 2 ) using Communicated by over 1000 landscape query images with ground truth GPS position.

Towards an Electronic Orientation Table: Using Features Extracted From the Image to Register Digital Elevation Model

“…In figure 7, we clearly see the relevance of each individual step, since the average distance dramatically increases (from 3 pixels to 14 or even 23) when we remove one of them. It is worth mentioning that our learning step has been run on the CH1 dataset described in (Saurer et al, 2016) and not on our dataset. The CH1 dataset consists of images of mountains in which the skyline has been manually segmented (unfortunately, no geographic information are provided with this dataset, so that we can not extract the correct DEM to run our algorithm on these images).…”

“…The approach of Saurer et al is a bit different from the previous ones (Saurer et al, 2016) because they classify all the pixels of the image into sky and non-sky pixels. The feature vector characterizing each pixel is a concatenation of 4 bags of words (textons, local ternary patterns, self-similarities and SIFT), each one being quantized to 512 clusters and independently in 5 different color spaces.…”

“…The aim of the work of Saurer et al is to geo-localize an image that has been acquired in the Alps (Saurer et al, 2016). So, from an image, they first extract the skyline as described below and try to find the most similar skyline in the DEM data they have.…”

VLASE: Vehicle Localization by Aggregating Semantic Edges

“…Based on human understandability, we broadly classify the localization features into the following two categories: Simple Features: We refer to simple features as the ones that are human-understandable: line-segments, horizon, road maps, and skylines. Skylines or horizon separating sky from buildings or mountains can be used for localiation [7]- [10]. Several existing methods use 3D models and/or omnidirectional cameras for geolocalization [8], [18]- [25].…”

Comparison of semantic segmentation approaches for horizon/sky line detection

“…This paper provides a quantitative comparison of four such methods for autonomous horizon/sky line detection on an extensive data set. Specifically, we provide the comparison between four recently proposed segmentation methods; one explicitly targeting the problem of horizon detection [2], second focused on visual geolocalization but relying on accurate detection of skyline [15] and other two proposed for general semantic segmentation -Fully Convolutional Networks (FCN) [21] and SegNet [22]. Each of the first two methods is trained on a common training set [11] comprised of about 200 images while models for the third and fourth method are fine tuned for sky segmentation problem through transfer learning using the same data set.…”