Land Cover Classification in Combined Elevation and Optical Images Supported by OSM Data, Mixed-level Features, and Non-local Optimization Algorithms

Bulatov, Dimitri; Häufel, Gisela; Lucks, Lukas; Pohl, Melanie

doi:10.14358/pers.85.3.179

Cited by 6 publications

(6 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Building and vegetation masks can be generated using land cover classification tasks based on the combination of spatial and spectral features derived from geodata. [6][7][8] Due to the fact that the focus of this work is based on the integration of the cover map into the path planning algorithm of the walking robot and its navigation, we accessed the existing vector data for the building mask using OpenStreetMap data. If aerial imagery (see Figure 3, lower right) was also available for the area, a vegetation mask could be derived using land cover classification.…”

Section: Test Environment -Data Pre-processingmentioning

confidence: 99%

See 1 more Smart Citation

Ghost: getting invisibly from position A to position B

Häufel,

Bulatov,

Emter

et al. 2023

Emerging Imaging and Sensing Technologies for Security and Defence VIII

View full text Add to dashboard Cite

The Ghost GR Vision 60 V5 robot is a walking robot from Ghost Robotics, which has extensive sensor and autonomy capabilities. The autonomy functions are provided by Fraunhofer IOSB’s Algorithm Toolbox (ATB) and enable the robot to perform localization, obstacle mapping, path planning, and path control. While Ghost autonomously moves from position A to position B, the environment is captured by laser scanners. Normally, the shortest path between the two points is taken. But what happens if this path can be seen by enemy observers and the walking robot is disturbed or, in the worst case, destroyed? At Fraunhofer IOSB, a fusion of the sensed environment based on laser scanners, existing height models and a coverage map was implemented on the walking robot’s system. This enables the robot to navigate tactically: It will – whenever possible – actively and autonomously seek cover in the shelter of houses and other objects on its way to the given target point.

show abstract

Section: Test Environment -Data Pre-processingmentioning

confidence: 99%

“…If aerial imagery (see Figure 3, lower right) was also available for the area, a vegetation mask could be derived using land cover classification. 6 In the following step, building and vegetation masks are composed.…”

Section: Test Environment -Data Pre-processingmentioning

confidence: 99%

Ghost: getting invisibly from position A to position B

Häufel,

Bulatov,

Emter

et al. 2023

Emerging Imaging and Sensing Technologies for Security and Defence VIII

View full text Add to dashboard Cite

show abstract

“…In this work, we use the smoothness assumption that neighboring pixels in the landcover map mostly have the same classes. This assumption as soft constraint (Schindler, 2012;Bulatov et al, 2019) yields the cost function…”

Section: Optimization On Markov Random Fieldmentioning

confidence: 99%

Identification of Misclassified Pixels in Semantic Segmentation With Uncertainty Evaluation

Budde

Bulatov

Iwaszczuk

2021

Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

Abstract. Classification, and in particular semantic segmentation, plays a major role in remote sensing. In remote sensing, the classes usually correspond to landcover or landuse types while the data elements are image pixels. The results are so-called semantically segmented pixels describing the content of the data for each pixel. The identification of misclassified pixels is essential to perceive the overall performance of the classification algorithm. In the case of semantic segmentation, it is typically done with ground truth labels. However, such ground truth labels are rare and mostly reserved for training only. Especially deep learning approaches are data-hungry algorithms requesting a lot of labeled examples. In this work, we explore the possibility of using Monte-Carlo dropout for the identification of model-induced misclassifications. In particular, we obtain uncertainty measures from several inferences induced by the Monte-Carlo dropout. Furthermore, we examine how Markov Random Field optimization can reduce the number of misclassifications and facilitate their identification. The extent to which uncertainties provide information about misclassifications is assessed. Our results allow detecting 51 % of the misclassifications using uncertainties. Application of Markov Random Field optimization leads to a reduction of the percentage of misclassifications while detecting 0.4 % more misclassifications as without.

show abstract

“…Sources of systematic errors in DTMs must be eliminated in order to avoid mis-classifications. Many authors (Bulatov et al, 2019, Häufel et al, 2018, Huang et al, 2015 presented approaches of training data acquisition for land cover classification. For example, in (Bulatov et al, 2019), segmentation results are assigned to one of four classes (building, grass, road, and tree) using cascades of simple features followed by an interactive verification step, which is not too time-consuming if only larger segments are considered.…”

Section: Motivationmentioning

confidence: 99%

“…Many authors (Bulatov et al, 2019, Häufel et al, 2018, Huang et al, 2015 presented approaches of training data acquisition for land cover classification. For example, in (Bulatov et al, 2019), segmentation results are assigned to one of four classes (building, grass, road, and tree) using cascades of simple features followed by an interactive verification step, which is not too time-consuming if only larger segments are considered. Among simple features used to differentiate between mentioned classes, especially relative elevation plays a crucial part.…”

Section: Motivationmentioning

confidence: 99%

DTM Correction in Areas of Steep Slopes

Häufel

Böge

Bulatov

2020

Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

Abstract. Computation of a DTM from a DSM is a well-known and very important task. We derive the DTM by a procedure consisting of ground points extraction, surface interpolation and triangulation by a canonical mesh if the terrain is flat or has only moderate changes in elevation. In regions with steep slopes, such as at riversides, and with man-made 3D structures, such as around bridges, interpolation artifacts and suppression of high-resolution details can lead to coarse errors in local elevations even for the building detection task. The eligible regions must be therefore detected and at least locally reprocessed. For detection, we search for connected components of a certain minimum size with negative relative elevations. For reconstruction, we suppress the points with erroneously reconstructed DSM values and interpolate the surface by means of L1 splines. Finally, these meshes must be fused into one single DTM mesh. We applied land cover classification to demonstrate the usability of our correction. The overall accuracy amounts to around 88% while the number of faulty assignments due to incorrect DTMs can be significantly reduced.

show abstract

Land Cover Classification in Combined Elevation and Optical Images Supported by OSM Data, Mixed-level Features, and Non-local Optimization Algorithms

Cited by 6 publications

References 0 publications

Ghost: getting invisibly from position A to position B

Ghost: getting invisibly from position A to position B

Identification of Misclassified Pixels in Semantic Segmentation With Uncertainty Evaluation

DTM Correction in Areas of Steep Slopes

Contact Info

Product

Resources

About