Aerial Stereo Images Research Articles

A Hierarchical Deformable Deep Neural Network and an Aerial Image Benchmark Dataset for Surface Multiview Stereo Reconstruction

Multiview stereo (MVS) aerial image depth estimation is a research frontier in the remote sensing field. Recent deep learning-based advances in close-range object reconstruction have suggested the great potential of this approach. Meanwhile, the deformation problem and the scale variation issue are also worthy of attention. These characteristics of aerial images limit the applicability of the current methods for aerial image depth estimation. Moreover, there are few available benchmark datasets for aerial image depth estimation. In this regard, this article describes a new benchmark dataset called the LuoJia-MVS dataset ( <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://irsip.whu.edu.cn/resources/resources_en_v2.php</uri> ), as well as a new deep neural network known as the hierarchical deformable cascade MVS network (HDC-MVSNet). The LuoJia-MVS dataset contains 7972 five-view images with a spatial resolution of 10 cm, pixel-wise depths, and precise camera parameters, and was generated from an accurate digital surface model (DSM) built from thousands of stereo aerial images. In the HDC-MVSNet network, a new full-scale feature pyramid extraction module, a hierarchical set of 3-D convolutional blocks, and “true 3-D” deformable 3-D convolutional layers are specifically designed by considering the aforementioned characteristics of aerial images. Overall and ablation experiments on the WHU and LuoJia-MVS datasets validated the superiority of HDC-MVSNet over the current state-of-the-art MVS depth estimation methods and confirmed that the newly built dataset can provide an effective benchmark.

Создание пространственной модели рельефа местности по результатам геодезических измерений для анализа воздействия на окружающую среду в условиях малых карьеров

Mining, especially open pit mining, has a significant impact on the terrain. The article discusses the use of geodetic methods to monitor the process of mining and its possible impact on the environ-ment. The created relief model can be used to design the process of quarry restoration after their closure. The mineral wealth of a country is considered an important factor and is linked to the country's economy. Mineral resources are non-renewable, and their extraction and subsequent processing must take into account the rarity of their occurrence, the level of mining technologies to track spatial changes in mining operations. The main idea of this scientific study is to apply geodetic measurement methods to determine the current scale of mining in a quarry using a 3D terrain model. The paper presents an overview of the technologies used in performing geodetic measurements, as well as processing the measured data to assess the state of the spatial distribution of a quarry in order to determine the impact of mining on the landscape. The purpose of this article is to create a spatial terrain model based on the results of geodetic measurements to analyze the impact on the environment in small open pits. Methods of geodetic observations were used to create a spatial model of the terrain. Digital maps were also created by photogrammetric processing of aerial stereo images and subsequent semi-automatic removal of vegetation and other objects.

ON THE EFFECT OF DEM QUALITY FOR LANDSLIDE SUSCEPTIBILITY MAPPING

Abstract. Generating precise and up-to-date landslide susceptibility maps (LSMs) in landslide-prone areas is important to identify hazard potential in the future. The data quality and the method selection affect the accuracy of the LSMs. In this context, the accuracy and precision of the digital elevation models (DEMs) used as input are among the most important performance elements. Therefore, the influence of DEM accuracy and spatial resolution in producing LSMs was investigated here. A high accuracy DEM with 5 m grid spacing produced from aerial photographs and the EU-DEM v1.1 freely accessible from Copernicus Land Monitoring Service with 25 m spatial resolution were used for producing two different LSMs using the Random Forest (RF) method in this study. The RF method has proven success for this purpose. A total of eight conditioning factors, which include topographical and geological features, was used as model input. The landslide inventory was derived with the help of aerial stereo images with 20 cm and 30 cm ground sampling distances. The performances of the LSMs were assessed with receiver operating characteristics (ROC) area under curve (AUC) values. In addition, the results were compared with visual inspection. The results show that although the AUC values obtained from the aerial DEM (0.95) and EU-DEM v1.1 (0.93) were comparable; based on the visual assessments, the LSM obtained from the higher resolution DEM was found more successful in detecting the landslides and thus exhibited better prediction performance.

Predicting the location of larval habitats of Anopheles mosquitoes using remote sensing and soil type data

Malaria is a mosquito-borne infectious disease transmitted by the bite of Anopheles mosquitoes. Accurate and timely identification of Anopheles larval habitats and analysis of environmental factors affecting the formation and stability of these locations are very effective in the prevention and spread of malaria. In the absence of sufficient field observations and environmental parameters required in a suitable spatial coverage, remote sensing data can be effective in predicting the habitats of Anopheles mosquitoes. In this article, high-risk depressions that have the potential for Anopheles larval habitats had been identified by fusing a Digital Surface Model (DSM) extracted from very high spatial resolution aerial stereo images with land-use and soil type maps. land-useFinally, the high-risk map of malaria based on the prone larval habitats of Anopheles was created. To evaluate the results, 22 important depressions were identified using field observations in the 1.5 km buffer around Qaleh-Ganj city, Kerman Province, Iran. Of these 22 samples, 14 of them were stable for more than one month at temperatures above 30 °C and the rest were able to store water for three to four weeks. 12 out of 14 samples were consistent with the identified 130 high-risk depressions in the buffer range. Also, 19 of these 22 samples were compatible with the optimal depressions in the buffer range. By comparing the proposed method with previous methods based on data with medium and low spatial resolution or meteorological data, it was concluded that the correct and accurate seasonal and temporary position of Anopheles larval habitats depend on higher spatial resolution of remote sensing data. In addition, this study demonstrated that the use of vegetation and water indices cannot predict the exact location of all habitats of Anopheles. Because many of these habitats were temporary and these indices could not estimate and predict their exact location. The proposed method can be applied to search for suitable larval habitats of Anopheles around local residential areas, where neither medium and low spatial resolution data nor limited field observations can be used.

COMPARISON OF SUBAQUATIC DIGITAL ELEVATION MODELS FROM AIRBORNE LASER SCANNING AND IMAGERY

Abstract. The paper describes and compares the workflows and results of generating digital elevation models (DEMs) of underwater areas from airborne laser scanning and aerial stereo images. Based on a combined laser scanning/image data set of an artificial lake, both methods are described and pros/cons are highlighted. The authors focus on the final results, especially on accuracy, completeness and spatial resolution of the underwater DEM’s. Further, practical aspects of processing and complexity of both methods are highlighted too.

Can regional aerial images from orthophoto surveys produce high quality photogrammetric Canopy Height Model? A single tree approach in Western Europe

Forest monitoring tools are needed to promote effective and data driven forest management and forest policies. Remote sensing techniques can increase the speed and the cost-efficiency of the forest monitoring as well as large scale mapping of forest attribute (wall-to-wall approach). Digital Aerial Photogrammetry (DAP) is a common cost-effective alternative to airborne laser scanning (ALS) which can be based on aerial photos routinely acquired for general base maps. DAP based on such pre-existing dataset can be a cost effective source of large scale 3D data. In the context of forest characterization, when a quality Digital Terrain Model (DTM) is available, DAP can produce photogrammetric Canopy Height Model (pCHM) which describes the tree canopy height. While this potential seems pretty obvious, few studies have investigated the quality of regional pCHM based on aerial stereo images acquired by standard official aerial surveys. Our study proposes to evaluate the quality of pCHM individual tree height estimates based on raw images acquired following such protocol using a reference filed-measured tree height database. To further ensure the replicability of the approach, the pCHM tree height estimates benchmarking only relied on public forest inventory (FI) information and the photogrammetric protocol was based on low-cost and widely used photogrammetric software. Moreover, our study investigates the relationship between the pCHM tree height estimates based on the neighboring forest parameter provided by the FI program.Our results highlight the good agreement of tree height estimates provided by pCHM using DAP with both field measured and ALS tree height data. In terms of tree height modeling, our pCHM approach reached similar results than the same modeling strategy applied to ALS tree height estimates. Our study also identified some of the drivers of the pCHM tree height estimate error and found forest parameters like tree size (diameter at breast height) and tree type (evergreenness/deciduousness) as well as the terrain topography (slope) to be of higher importance than image survey parameters like the variation of the overlap or the sunlight condition in our dataset. In combination with the pCHM tree height estimate, the terrain slope, the Diameter at Breast Height (DBH) and the evergreenness factor were used to fit a multivariate model predicting the field measured tree height. This model presented better performance than the model linking the pCHM estimates to the field tree height estimates in terms of r² (0.90 VS 0.87) and root mean square error (RMSE, 1.78 VS 2.01 m). Such aspects are poorly addressed in literature and further research should focus on how pCHM approaches could integrate them to improve forest characterization using DAP and pCHM. Our promising results can be used to encourage the use of regional aerial orthophoto surveys archive to produce large scale quality tree height data at very low additional costs, notably in the context of updating national forest inventory programs.

FORests and HYdrology under Climate Change in Switzerland v1.0: a spatially distributed model combining hydrology and forest dynamics

Abstract. We present FORHYCS (FORests and HYdrology under Climate Change in Switzerland), a distributed ecohydrological model to assess the impact of climate change on water resources and forest dynamics. FORHYCS is based on the coupling of the hydrological model PREVAH and the forest landscape model TreeMig. In a coupled simulation, both original models are executed simultaneously and exchange information through shared variables. The simulated canopy structure is summarized by the leaf area index (LAI), which affects local water balance calculations. On the other hand, an annual drought index is obtained from daily simulated potential and actual transpiration. This drought index affects tree growth and mortality, as well as a species-specific tree height limitation. The effective rooting depth is simulated as a function of climate, soil, and simulated above-ground vegetation structure. Other interface variables include stomatal resistance and leaf phenology. Case study simulations with the model were performed in the Navizence catchment in the Swiss Central Alps, with a sharp elevational gradient and climatic conditions ranging from dry inner-alpine to high alpine. In a first experiment, the model was run for 500 years with different configurations. The results were compared against observations of vegetation properties from national forest inventories, remotely sensed LAI, and high-resolution canopy height maps from stereo aerial images. Two new metrics are proposed for a quantitative comparison of observed and simulated canopy structure. In a second experiment, the model was run for 130 years under climate change scenarios using both idealized temperature and precipitation change and meteorological forcing from downscaled GCM-RCM model chains. The first experiment showed that model configuration greatly influences simulated vegetation structure. In particular, simulations where height limitation was dependent on environmental stress showed a much better fit to canopy height observations. Spatial patterns of simulated LAI were more realistic than for uncoupled simulations of the forest landscape model, although some model deficiencies are still evident. Under idealized climate change scenarios, the effect of the coupling varied regionally, with the greatest effects on simulated streamflow (up to 60 mm yr−1 difference with respect to a simulation with static vegetation parameters) seen at the valley bottom and in regions currently above the treeline. This case study shows the importance of coupling hydrology and vegetation dynamics to simulate the impact of climate change on ecosystems. Nevertheless, it also highlights some challenges of ecohydrological modeling, such as the need to realistically simulate the plant response to increased CO2 concentrations and process uncertainty regarding future land cover changes.

Utility of image point cloud data towards generating enhanced multitemporal multisensor land cover maps

Multitemporal land cover classification over urban areas is challenging, especially when using heterogeneous data sources with variable quality attributes. A prominent challenge is that classes with similar spectral signatures (such as trees and grass) tend to be confused with one another. In this paper, we evaluate the efficacy of image point cloud (IPC) data combined with suitable Bayesian analysis based time-series rectification techniques to improve the classification accuracy in a multitemporal context. The proposed method uses hidden Markov models (HMMs) to rectify land covers that are initially classified by a random forest (RF) algorithm. This land cover classification method is tested using time series of remote sensing data from a heterogeneous and rapidly changing urban landscape (Kuopio city, Finland) observed from 2006 to 2014. The data consisted of aerial images (5 years), Landsat data (all 9 years) and airborne laser scanning data (1 year). The results of the study demonstrate that the addition of three-dimensional image point cloud data derived from aerial stereo images as predictor variables improved overall classification accuracy, around three percentage points. Additionally, HMM-based post processing reduces significantly the number of spurious year-to-year changes. Using a set of 240 validation points, we estimated that this step improved overall classification accuracy by around 3.0 percentage points, and up to 6 to 10 percentage points for some classes. The overall accuracy of the final product was 91% (kappa = 0.88). Our analysis shows that around 1.9% of the area around Kuopio city, representing a total area of approximately 0.61 km2, experienced changes in land cover over the nine years considered.

Comparison of multispectral airborne laser scanning and stereo matching of aerial images as a single sensor solution to forest inventories by tree species

Airborne Light Detection and Ranging (LiDAR) information alone is insufficient for species-specific prediction of forest stand attributes, and therefore auxiliary optical image features (OIF) are commonly used to decrease the prediction errors associated with species-specific tree attributes. However, this requires collection and merging of two data sources, LiDAR and OIF, which increases the costs of the inventory. The recently introduced multispectral LiDAR (M-ALS) provides a potential single-sensor solution for obtaining species-specific information, as its multispectral intensity values resemble optical image data. Image point clouds (IPC) derived from aerial stereo images are another single-sensor option that provides both geometric and optical information. We compared two single-sensor options, M-ALS and IPC, with two LiDAR data sets (leaf-on and leaf-off) with auxiliary OIF, for the prediction of boreal tree species' volumes. In terms of root-mean-square error (RMSE) in the validation data, the LiDAR+OIF combination performed best (leaf-on RMSE: 33.3%; leaf-off RMSE 34.3%), followed by M-ALS + OIF (RMSE = 35.2%) and IPC + OIF (RMSE = 42.4%). The mean RMSE value associated with M-ALS increased to the same level (44.7%) as the IPC + OIF combination when optical image features were not included. Both IPC and M-ALS are potential single sensor solutions for forest inventories, but the use of both LiDAR and OIF provides the most accurate results.

GEOMETRIC OBJECT BASED BUILDING RECONSTRUCTION FROM SATELLITE IMAGERY DERIVED POINT CLOUDS

Abstract. 3D building models are needed for urban planning and smart city. These models can be generated from stereo aerial images, satellite images or LiDAR point clouds. In this paper, we propose a geometric object-based building reconstruction method from satellite imagery derived point clouds. The goal is to achieve a geometrically correct, topologically consistent, and non-redundant 3D representation for buildings in urban areas. The paper first introduces our motivation, followed by a comprehensive review on related works. We then introduce the methodology and process developed in this paper. Primary results from the point clouds generated from WorldView high resolution satellite images are used to demonstrate the performance of the approach.

A Framework for the Long-term Monitoring of Urban Green Volume Based on Multi-temporal and Multi-sensoral Remote Sensing Data

Green urban infrastructure is of key importance for many aspects of urban life and urban planning. Valid and comprehensive databases with very high spatial and temporal resolution are needed to monitor changes and to detect negative trends. This paper presents an approach to assess urban indicators such as green volume and soil sealing with very high accuracy and based on a wide range of different sensors (aerial stereo images, QuickBird, WorldView 2 and 3, Sentinel 2, HRSC, LIDAR). A framework using regression tree methods was developed and successfully applied in a case study (the city of Potsdam, Germany) resulting in a long time series dating back 25 years. The methodology offers the opportunity to analyze urban development in detail and to understand the functional relationships of urban planning processes. Demands for effective climate change adaptation, especially in terms of reducing heat stress, can thus be better defined.

A Relative Rotation between Two Overlapping UAV's Images

In this paper, we study the influence of varying baseline components on the accuracy of a relative rotation between two overlapping aerial images taken form UAV flight. The case is relevant when mosaicking UAV's aerial images by registering each individual image. Geotagged images facilitated by a navigational grade GPS receiver on board inform the camera position when taking pictures. However, these low accuracies of geographical coordinates encoded in an EXIF format are unreliable to depict baseline vector components between subsequent overlapping images. This research investigates these influences on the stability of rotation elements when the vector components are entered into a standard coplanarity condition equation to determine the relative rotation of the stereo images. Assuming a nadir looking camera on board while the UAV platform is flying at a constant height, the resulted vector directions are utilized to constraint the coplanarity equation. A detailed analysis of each variation is given. Our experiments based on real datasets confirm that the relative rotation between two successive overlapping image is practically unaffected by the accuracy of positioning method. Furthermore, the coplanarity constraint is invariant with respect to a translation along the baseline of the aerial stereo images.

EVALUATION OF DEEP LEARNING BASED STEREO MATCHING METHODS: FROM GROUND TO AERIAL IMAGES

Abstract. Dense stereo matching has been extensively studied in photogrammetry and computer vision. In this paper we evaluate the application of deep learning based stereo methods, which were raised from 2016 and rapidly spread, on aerial stereos other than ground images that are commonly used in computer vision community. Two popular methods are evaluated. One learns matching cost with a convolutional neural network (known as MC-CNN); the other produces a disparity map in an end-to-end manner by utilizing both geometry and context (known as GC-net). First, we evaluate the performance of the deep learning based methods for aerial stereo images by a direct model reuse. The models pre-trained on KITTI 2012, KITTI 2015 and Driving datasets separately, are directly applied to three aerial datasets. We also give the results of direct training on target aerial datasets. Second, the deep learning based methods are compared to the classic stereo matching method, Semi-Global Matching(SGM), and a photogrammetric software, SURE, on the same aerial datasets. Third, transfer learning strategy is introduced to aerial image matching based on the assumption of a few target samples available for model fine tuning. It experimentally proved that the conventional methods and the deep learning based methods performed similarly, and the latter had greater potential to be explored.

Comparative accuracy of the AW3D30 DSM, ASTER GDEM, and SRTM1 DEM: A case study on the Zaoksky testing ground, Central European Russia

ABSTRACTQuasi-global, medium-resolution land surface models – the Shuttle Radar Topography Mission digital elevation model (SRTM1 DEM), the Advanced Spaceborne Thermal Emission and Reflection Radiometer Global DEM (ASTER GDEM), and the Advanced Land Observing Satellite World 3D digital surface model (AW3D30 DSM) – are widely used in geosciences. We carried out a comparative analysis of accuracy for these models exemplified by the Zaoksky testing ground, Central European Russia. First, we extracted related portions from the SRTM1 DEM, ASTER GDEM, and AW3D30 DSM for the study area. In the further analysis, we used 69 ground control points (GCPs) within the testing ground, as well as a reference DEM photogrammetrically generated from aerial stereo images. For each GCP, we determined an elevation value in four models and a related vertical error. For each model, we carried out a statistical analysis of vertical errors. To visualize the spatial distribution of vertical errors, we calculated matrices of residuals between the reference DEM and the quasi-global models. The results demonstrate that the AW3D30 DSM can be considered the most accurate quasi-global, medium-resolution land surface model. We do not recommend the use of the ASTER GDEM because it is marked by high errors with chaotically distributed hummock-like patterns.

Stereo-imagery-based post-stratification by regression-tree modelling in Swiss National Forest Inventory

Swiss National Forest Inventory (NFI) produces estimates of characteristics related to the state and change of the forest resources in the whole country as well as in smaller regions and domains. The estimates are based on measurements on permanent field plots placed on a systematic grid over the country. Digital stereo aerial images are an important source of auxiliary information that can be used with post-stratified estimation to increase the precision of the estimates. We examined post-stratification based on stereo image information, available on a much denser grid than the field plots, for the estimation of forest area, total growing-stock volume and growing-stock volume per unit of accessible forest area. We considered 31 post-stratification schemes that consisted of (1) forest/non-forest stratification by four alternative methods based on manual or semi-automated stereo image interpretation, and (2) sub-stratification within the forest stratum by regression-tree models constructed between field-measured growing-stock volume and stereo-image-retrieved vegetation height characteristics. We applied the schemes in two real NFI estimation cycles, and found that the most precise schemes reduced the variance of the whole-country estimators of forest area, total growing-stock volume and growing-stock volume per unit of accessible forest area by 81%, 67% and 34%, respectively, compared to the estimator without post-stratification. Based on this study, we consider regression-tree modelling well-suited for post-stratification, as it produces stratum boundaries as a direct result of modelling, appears robust against deviating values in fitting data, and avoids problems with unrealistic model predictions in application.

Forest Change Detection by Using Point Clouds From Dense Image Matching Together With a LiDAR-Derived Terrain Model

Recent development of image matching algorithms has enabled the generation of highly detailed image point clouds (IPCs) from aerial stereo images. The acquisition of aerial images is typically significantly less expensive than the collection of light detection and ranging (LiDAR) or field data. IPCs thus provide a cost-effective alternative to update forest resource data more frequently than at present. We evaluate the feasibility of IPCs in the detection of forest changes by assessing the accuracy of automatic IPC-based classification of thinnings and clear cuts. IPCs were created by using the semi-global matching and next-generation automatic terrain extraction algorithms. To predict changes over a period of three years, we created difference layers which displayed the difference in height or volume between the initial and subsequent time points. These were constructed by calculating the difference in either IPC-based canopy height models or in IPC-derived height and volume models. In this process, the LiDAR-derived digital terrain model was used to scale heights to above ground level. The values from the difference layers were then used in logistic regression models to classify the study area into the categories ClearCut, Thinning, or NoChange. When the predicted changes were compared with the true changes verified in the field, we obtained a classification accuracy for clear cuts 98.6% at best, but only 24.1% for thinnings. In conclusion, IPCs are applicable for the detection of major changes in forests, but incapable of detecting minor changes. The same seems to be applied to LiDAR data.

Automatic 3D Coordinate Extraction from High Resolution Digital Aerial Images

Image matching has been widely used in various fields such as digital photogrammetry, remote sensing, image processing and computer vision. In this study, the speeded up robust feature extraction algorithm for automatic point extraction was applied to a pair of high resolution digital aerial stereo images. In contrast to other studies, the universal transverse mercator coordinates of randomly selected test points were computed automatically with the proposed approach and compared with the same points that were manually measured. As a result, the mean square differences for the UTM coordinates’ in the X, Y and Z axes were calculated as ±0.043, 0.030 and 0.153 m, respectively. It is proved that the proposed algorithm is successful in matching remotely sensed high-resolution images acquired with digital cameras. Moreover, the automatization of the basic tasks in image matching and the 3D coordinate acquisition were also achieved by the algorithm.

Line matching using a disparity map in rectified image space for stereo aerial images

ABSTRACTThis study proposes a novel line-matching method that uses a disparity map in rectified image space for stereo aerial images. Two original images are first rectified to images in the rectified image space by a rectifying projection. The disparity map is then built by dense matching technology in the rectified image space. For an unmatched line segment, the surrounding pixels on the disparity map can be triangulated into 3D points, which can be used to compute rectified planar homography and obtain a predicted line segment. The candidate matches are determined by calculating line similarity as measured by position and orientation shifts with respect to the predicted line segment. Three pairs of calibrated aerial images were used to demonstrate the efficiency and validate the proposed method. Compared with another line-matching method using sparse image points, the number of matches, accuracy and completeness are improved in the proposed method.

Das aktuelle Vegetationshöhenmodell der Schweiz: spezifische Anwendungen im Waldbereich

Current model of canopy heights in Switzerland: specific applications in forestry A national vegetation height model was calculated for Switzerland for the first time last year using digital aerial images. The ADS80 stereo aerial images, which were routinely gathered by the Swiss Federal Office of Topography (swisstopo), contain information about the height of vegetation in forests. We used the stereo aerial images to first calculate a digital surface model (DSM) with a very high spatial resolution (1 × 1 m). The DSM was then normalized to obtain the actual vegetation heights using a digital terrain model (DTM) based on laser data with the buildings masked out, and to produce a vegetation height model (VHM). Such a model could be calculated in the framework of the Swiss National Forest Inventory (NFI) with consistent methods and a very high level of detail. For covering the whole of Switzerland, we used summer aerial images from the years 2007 to 2012. The VHM reached almost nationwide coverage (98% of the country's surface area). Some areas, such as steep mountainsides or very bright glaciers, were problematic for calculating the model, and appear in it as gaps. Vegetation height information collected with this method is most useful for analyzing entire forest stands, but the data do not have a high enough spatial resolution for single-tree-based analyses. The VHM can be applied in a wide variety of ways. Here, we describe three of these: 1) generating forest masks, 2) extracting forest canopy gaps, and 3) detecting changes in the stocking of a forested area.

Quality assessment of high density digital surface model over different land cover classes

Background and Purpose: Recent research on generation of digital surface models (DSMs) using image matching methods revealed a great potential of DSM application in forestry, especially in forest inventory. However, research dealing with DSM generation from digital aerial images are still lacking in Croatia. Therefore, the main objective of this study was to present the workflow for generating high density DSM from colour infrared (CIR) digital stereo aerial images using area-based image matching algorithm. Materials and Methods: The high density DSM was generated from colour infrared digital aerial stereo images using Dense DTM algorithm of PHOTOMOD software - an area-based image matching algorithm which operates on the principle of cross-correlation approach. To evaluate the quality of the generated DSM, an agreement assessment with manual stereo measurements was conducted over three different land cover classes (forests, shrubs, grasslands) using the same images as for DSM generation. Results: The good vertical agreement between the generated DSM and stereo measurement was achieved for all three land cover classes present at the research area. The highest vertical agreement was obtained for the grassland land cover class (RMSE=0.36), slightly lower for forest (RMSE=0.62), whereas the lowest vertical agreement was obtained for shrub land cover class (RMSE=0.83). Conclusions: The results of this research are very promising and suggest that the high density DSM generated from digital aerial stereo images and by using the proposed methodology has the potential to be used in forestry, primarily in forest inventory. Therefore, further research should be focused on generation of CHM by subtracting available DTM from the high density DSM and on the examination of its potential for deriving various forest attributes.