Drift Reduction in Terrestrial Laser Scanning via Linear Dual Quaternion Interpolation

An automated method to register airborne and terrestrial laser scanning point clouds

HighlightsDual quaternion interpolation works better than the least squares method or decoupled interpolation of rotations and translations to distribute the closure error of 3D point cloud registration circuits.What are the main findings?Linear dual quaternion interpolation is the most efficient solution for distributing the closure error along a circuit of 3D poses.Our CSI (Constant Smooth Interpolation) method unifies the translation and rotation correction of an entire circuit of poses in a single step.What are the implications of the main findings?Easy implementation. It is possible to correct the closure error of the circuit just by linearly interpolating eight parameters.The interpolation can be performed fast and without losing the property of the shortest path on the manifold of the 3D poses.Laser scanning allows for the rapid acquisition of three-dimensional data in the form of 3D point clouds. However, due to the accumulation of errors in the registration of multiple pairs of point clouds along the sensor’s trajectory, the generated 3D reconstructions exhibit drift, which creates global inconsistencies in the scan. To address this error, there are drift correction models that distribute the error along a closed circuit of stations. In this work, we present a model of this nature based on the linear interpolation of dual quaternions. This linear solution simultaneously refines rotations and translations in a closed trajectory without iterative computations or matrix decomposition. Experimental evaluations on eight TLS datasets indicate that the proposed drift correction model provides a robust average error reduction of 26%, with a maximum reduction of 41% in circuits with large drift. This simultaneous solution improves pose accuracy in closed trajectories with theoretical advantages that translate into efficient and fast implementation. Although validated using TLS data, the proposed pose-circuit correction model is sensor-agnostic and can be applied to other 3D mapping systems.

The alpine landscape is rugged, dominated by glacial morphogenesis, comprising specific landforms of different sizes and shapes with a marked vertical relief and hierarchical-ordering of the forms. Digital geomorphometric analyses of such a land surface in a high level of detail can exploit mainly data acquired by airborne laser scanning or photogrammetry. However, the level of detail captured is limited to several meters or decimetres in these datasets for the relatively high above ground flying heights. Terrestrial laser scanning (TLS) and close range photogrammetry generate 3-D point clouds of ultra-high spatial resolution but their application is limited by extreme environmental conditions. TLS data often contain data shadows in alpine landscape resulting in inhomogeneous spatial distribution of the acquired 3-D point cloud. Such data properties cause creation of overly smoothed surfaces or artefacts in digital elevation models (DEMs) interpolated into a grid (raster). The sub-horizontal field of view in TLS can be compensated by sub-vertical field of view of digital cameras installed on unpiloted aerial platforms (UAVs). Such a photogrammetric 3-D reconstruction of terrain with UAVs is based on structure-from-motion (SfM) image matching techniques. The measurement precision and accuracy of UAV-SfM is lower than with TLS but the UAV-SfM data can be used in filling the TLS data voids. In this paper, we present the results of combined use of TLS and UAV-SfM for high-resolution modelling of a glacial cirque in the Tatry Mountains, the Carpathians, Europe. The achieved accuracy (1 standard deviation) of mutual co-registration of 18 TLS positions was 4.2 mm. The accuracy of georeferencing the final TLS data in the national cartographic system was 33.2 mm based on 15 ground control points. The UAV-SfM dataset was spatially co-registered on the TLS dataset with the accuracy of 137 mm and point filling the TLS voids were used to generate the final DEM of 0.5m cell size from the combined point clouds.

With the increasing use of terrestrial laser scanning (TLS) technology in the field of forest ecology, a large number of studies have been carried out on the separation of wood and leaves based on TLS point cloud data. However, most wood–leaf separation methods adopt the point‐wise classification strategy, which is not efficient for processing large‐volume TLS datasets acquired at the forest plot level. In this study, we proposed a segment‐wise classification strategy to improve the efficiency of the wood–leaf separation from large‐volume TLS point cloud datasets collected at the forest plot. The proposed method first decomposes the point cloud into three parts based on the threshold values of its local curvature. Then, the first two parts with lower local curvatures were segmented respectively by a connected component labelling algorithm. Finally, the segmented point clouds were classified into wood or leaf segments according to the segment‐wise geometric features of each segment. We tested our method on both needleleaf and broadleaf forest plots in temperate and tropical forests. We also compared our method with two other state‐of‐the‐art wood–leaf separation methods, that is, the CANUPO and LeWoS. The results showed that our method was more than 10 times faster than the compared methods while maintaining comparable and even higher accuracy. Our study demonstrates that the segment‐wise classification strategy applies to the large‐volume TLS datasets and can greatly improve the efficiency of the classification. The proposed method is simple, fast and universally applicable to the TLS data from various tree species and forest types at the plot level, which may facilitate the adoption of TLS technology by forest ecologists in their studies.

Landslides are one of the critical natural hazards that cause human, infrastructure, and economic losses. Risk of catastrophic losses due to landslides is significant given sprawled urban development near steep slopes and the increasing proximity of large populations to hilly areas. For reducing these losses, a high-resolution digital terrain model (DTM) is an essential piece of data for a qualitative or a quantitative investigation of slopes that may lead to landslides. Data acquired by a terrestrial laser scanning (TLS), called a point cloud, has been widely used to generate a DTM, since a TLS is appropriate for detecting small- to large-scale ground features on steep slopes. For an accurate DTM, TLS data should be filtered to remove non-ground points, but most current algorithms for extracting ground points from a point cloud have been developed for airborne laser scanning (ALS) data and not TLS data. Moreover, it is a challenging task to generate an accurate DTM from a steep-slope area by using existing algorithms. For these reasons, we developed an algorithm to automatically extract only ground points from the point clouds of steep terrains. Our methodology is focused on TLS datasets and utilizes the adaptive principal component analysis–triangular irregular network (PCA-TIN) approach. Our method was applied to two test areas and the results showed that the algorithm can cope well with steep slopes, giving an accurate surface model compared to conventional algorithms. Total accuracy values of the generated DTMs in the form of root mean squared errors are 1.84 cm and 2.13 cm over the areas of 5252 m2 and 1378 m2, respectively. The slope-based adaptive PCA-TIN method demonstrates great potential for TLS-derived DTM construction in steep-slope landscapes.

Among emerging 3D scanning and imaging techniques that are commercially available, simultaneous localization and mapping (SLAM) is being substantially studied to generate 2D/3D maps of an unknown environment while reliably keeping track of the user’s pinpoint locations. Its ubiquitous mobility has demonstrated great mapping capabilities for infrastructures where vertical information may frequently be occluded using unmanned aircraft system (UAS) structure from-motion (SfM) photogrammetry. In addition, indoor mapping with terrestrial laser scanning can be a cumbersome task due to possible multiple scan locations. Intending to provide a cohesive 3D model by fusing point clouds collected via aerial SfM photogrammetry, terrestrial laser scanning (TLS), and SLAM, the purpose of this work is to assess the performance of SLAM point cloud generated by a proprietary mobile backpack laser scanner (BLS). Considering maximum scanning range and information integration strategy as variables, the point clouds generated by the BLS were evaluated against SfM and TLS datasets in terms of the internal consistency as well as external accuracy. TLS, SfM and SAM data collection efforts were made in a typical university campus environment. For the internal consistency, the SLAM-based point cloud with a maximum scanning range of 70 m presented a root mean square error (RMSE) of 2 mm. The SLAM+GNSS-based point cloud presented the lowest internal precision of RMSE = 0.861 m. The SLAM+GNSS 70 point cloud after a fine adjustment of misalignment presented the highest vertical accuracy with an RMSE = 0.069 m, while the point cloud generated from SfM photogrammetry presented RMSE = 0.297 m. The BLS was able to generate point cloud with an accuracy similar to GNSS-RTK surveying and it can be considered as a viable solution for indoor and outdoor mapping applications.

The fully automatic registration of 3D terrestrial laser scanning (TLS) point clouds, which is the first step in the usage of point clouds, is a highly challenging task in Light Detection and Ranging (LiDAR) remote sensing applications. Here, an automatic algorithm for pairwise coarse registration of TLS point clouds using 3D line features is proposed. First, the 3D line sets were extracted from the original pair of point clouds respectively, and two arbitrary lines from a point cloud were used to construct a 2-line base; Then, a pair of conjugate 2-line bases were identified from the source and target 3D line sets at a time, based on which a 3D rotation matrix together with its corresponding overlap between the pairwise 3D line sets was calculated; Third, a series of 3D rotation matrixes together with their overlaps were obtained using the traversing strategy to identify conjugate 2-line base pairs, and the 3D rotation matrix with the highest overlap was outputted as the final 3D rotation matrix, which was further used to compute the corresponding lines set from the pairwise 3D line sets; Next, based on the set of common perpendicular midpoints between the line correspondences, the 3D translation vector was computed; Finally, the 3D transformation matrix between pairwise point clouds was obtained by combining the 3D rotation matrix and the 3D translation vector. The proposed method was tested on three different TLS datasets, with experimental results demonstrating that the proposed algorithm could perform well on registering pairwise TLS point clouds. The rotation and translation errors of aligning the nine experimental pairwise point clouds were all less than 0.50° and 0.55m, respectively. This registration framework was also shown to be superior to state-of-the-art methods in terms of registration accuracy.

Calibration and validation of aboveground biomass (AGB) (AGB) products retrieved from satellite-borne sensors require accurate AGB estimates across hectare scales (1 to 100ha). Recent studies recommend making use of non-destructive terrestrial laser scanning (TLS) based techniques for individual tree AGB estimation that provide unbiased AGB predictors. However, applying these techniques across large sites and landscapes remains logistically challenging. Unoccupied aerial vehicle laser scanning (UAV-LS) has the potential to address this through the collection of high density point clouds across many hectares, but estimation of individual tree AGB based on these data has been challenging so far, especially in dense tropical canopies. In this study, we investigated how TLS and UAV-LS can be used for this purpose by testing different modelling strategies with data availability and modelling framework requirements. The study included data from four forested sites across three biomes: temperate, wet tropical, and tropical savanna. At each site, coincident TLS and UAV-LS campaigns were conducted. Diameter at breast height (DBH) and tree height were estimated from TLS point clouds. Individual tree AGB was estimated for ≥170 trees per site based on TLS tree point clouds and quantitative structure modelling (QSM), and treated as the best available, non-destructive estimate of AGB in the absence of direct, destructive measurements. Individual trees were automatically segmented from the UAV-LS point clouds using a shortest-path algorithm on the full 3D point cloud. Predictions were evaluated in terms of individual tree root mean square error (RMSE) and population bias, the latter being the absolute difference between total tree sample population TLS QSM estimated AGB and predicted AGB. The application of global allometric scaling models (ASM) at local scale and across data modalities, i.e., field-inventory and light detection and ranging LiDAR metrics, resulted in individual tree prediction errors in the range of reported studies, but relatively high population bias. The use of adjustment factors should be considered to translate between data modalities. When calibrating local models, DBH was confirmed as a strong predictor of AGB, and useful when scaling AGB estimates with field inventories. The combination of UAV-LS derived tree metrics with non-parametric modelling generally produced high individual tree RMSE, but very low population bias of ≤5% across sites starting from 55 training samples. UAV-LS has the potential to scale AGB estimates across hectares with reduced fieldwork time. Overall, this study contributes to the exploitation of TLS and UAV-LS for hectare scale, non-destructive AGB estimation relevant for the calibration and validation of space-borne missions targeting AGB estimation.

Error in target-based georeferencing and registration in terrestrial laser scanning

Terrestrial laser scanning (TLS) is a promising technology for quantity checking huge grain stocks with low cost, light workload and high efficiency. Existing applications of TLS in bulk grain measurement and quantification lack the ability to capture complete structural information of grain bulks and thus will result in large errors. In this paper, we propose a bi-temporal TLS scheme for fast 3D modeling and accurate volume measurement of bulk grains stored in large warehouses. The scheme uses bi-temporal multi-site TLS datasets acquired under both empty and full or high loading conditions to obtain complete surface information about grain bulk’s structure. In order for a grain bulk’s all external surfaces and the 3D volumetric model reconstructed therefrom to be automatically derived from the bi-temporal TLS dataset, several dedicated methods are developed for the scheme. A fully automated marker-free strategy exploring structurally semantic information inherent in the large grain storehouses is adopted to register multi-scan TLS point cloud data captured in large-scale granary scenes. Also, a local minima-based region growing technique is devised to extract underlying surfaces from a granary scene point cloud model. Experiments show that the proposed 3D modeling and volume measurement scheme can work effectively and quickly in TLS-based granary field applications and repeated test data demonstrate its correctness, repeatability and accuracy. Compared with the conventional manual measurement approach, the bi-temporal TLS scheme can not only achieve much higher measurement precision, but also greatly improve efficiency by significantly reducing cost, workload, and manpower. It has good potential for use in the area of nation-wide grain inventory inspection in China.

Terrestrial laser scanning (TLS) has become a powerful data acquisition technique for high-resolution high-accuracy topographic and morphological studies. Conventional static TLS surveys require setting up numerous reflectors (tie points) in the field for point clouds registration and georeferencing. To reduce surveying time and simplify field operational tasks, we have developed a rapid TLS surveying method that requires only one reflector in the field. The method allows direct georeferencing of point clouds from individual scans to an East–North–Height (ENH) coordinate system tied to a stable geodetic reference frame. TLS datasets collected at a segment of the beach–dune–wetland area in Freeport, Texas, USA are used to evaluate the performance of the rapid surveying method by comparing with kinematic GPS measurements. The rapid surveying method uses two GPS units mounted on the scanner and a reflector for calculating the northing angle of the scanner’s own coordinate system (SOCS). The Online Positioning User Service (OPUS) is recommended for GPS data processing. According to this study, OPUS Rapid-Static (OPUS-RS) solutions retain 1–2 cm root mean square (RMS) accuracy in the horizontal directions and 2–3 cm accuracy in the vertical direction for static observational sessions of approximately 30 min in the coastal region of Texas, USA. The rapid TLS surveys can achieve an elevation accuracy (RMS) of approximately 3–5 cm for georeferenced points and 2–3 cm for digital elevation models (DEMs). The elevation errors superimposed into the TLS surveying points roughly fit a normal distribution. The proposed TLS surveying method is particularly useful for morphological mapping over time in coastal regions, where strong wind and soft sand prohibit reflectors from remaining strictly stable for a long period. The theories and results presented in this paper are beneficial to researchers who frequently utilize TLS datasets in their research, but do not have opportunities to be involved in field data acquisition.

Rigid registration of 3D indoor scenes is a fundamental yet vital task in various fields that include remote sensing (e.g., 3D reconstruction of indoor scenes), photogrammetry measurement, geometry modeling, etc. Nevertheless, state-of-the-art registration approaches still have defects when dealing with low-quality indoor scene point clouds derived from consumer-grade RGB-D sensors. The major challenge is accurately extracting correspondences between a pair of low-quality point clouds when they contain considerable noise, outliers, or weak texture features. To solve the problem, we present a point cloud registration framework in view of RGB-D information. First, we propose a point normal filter for effectively removing noise and simultaneously maintaining sharp geometric features and smooth transition regions. Second, we design a correspondence extraction scheme based on a novel descriptor encoding textural and geometry information, which can robustly establish dense correspondences between a pair of low-quality point clouds. Finally, we propose a point-to-plane registration technology via a nonconvex regularizer, which can further diminish the influence of those false correspondences and produce an exact rigid transformation between a pair of point clouds. Compared to existing state-of-the-art techniques, intensive experimental results demonstrate that our registration framework is excellent visually and numerically, especially for dealing with low-quality indoor scenes.

Separating foliage and woody components can effectively improve the accuracy of simulating the forest eco-hydrological processes. It is still challenging to use deep learning models to classify canopy components from the point cloud data collected in forests by terrestrial laser scanning (TLS). In this study, we developed a convolution neural network (CNN)-based model to separate foliage and woody components (FWCNN) by combing the geometrical and laser return intensity (LRI) information of local point sets in TLS datasets. Meanwhile, we corrected the LRI information and proposed a contribution score evaluation method to objectively determine hyper-parameters (learning rate, batch size, and validation split rate) in the FWCNN model. Our results show that: (1) Correcting the LRI information could improve the overall classification accuracy (OA) of foliage and woody points in tested broadleaf (from 95.05% to 96.20%) and coniferous (from 93.46% to 94.98%) TLS datasets (Kappa ≥ 0.86). (2) Optimizing hyper-parameters was essential to enhance the running efficiency of the FWCNN model, and the determined hyper-parameter set was suitable to classify all tested TLS data. (3) The FWCNN model has great potential to classify TLS data in mixed forests with OA > 84.26% (Kappa ≥ 0.67). This work provides a foundation for retrieving the structural features of woody materials within the forest canopy.

Point cloud classification is quite challenging due to the influence of noise, occlusion, and the variety of types and sizes of objects. Currently, most methods mainly focus on subjectively designing and extracting features. However, the features rely on prior knowledge, and it is also difficult to accurately characterize the complex objects of point clouds. In this paper, we propose a concise multi-scale convolutional network (MSNet) for adaptive and robust point cloud classification. Both the local feature and global context are incorporated for this purpose. First, around each point, the spatial contexts of different sizes are partitioned as voxels of different scales. A voxel-based MSNet is then simultaneously applied at multiple scales to adaptively learn the discriminative local features. The class probability of a point cloud is predicted by fusing the features together across multiple scales. Finally, the predicted class probabilities of MSNet are optimized globally using the conditional random field (CRF) with a spatial consistency constraint. The proposed method was tested with data sets of mobile laser scanning (MLS), terrestrial laser scanning (TLS), and airborne laser scanning (ALS) point clouds. The experimental results show that the proposed method was able to achieve appreciable classification accuracies of 83.18%, 98.24%, and 97.02% on the MLS, TLS, and ALS data sets, respectively. The results also demonstrate that the proposed network has a strong generalization capability for classifying different kinds of point clouds under complex urban environments.

Interest in measuring forest biomass and carbon stock has increased as a result of the United Nations Framework Convention on Climate Change, and sustainable planning of forest resources is therefore essential. Biomass and carbon stock estimates are based on the large area estimates of growing stock volume provided by national forest inventories (NFIs). The estimates for growing stock volume based on the NFIs depend on stem volume estimates of individual trees. Data collection for formulating stem volume and biomass models is challenging, because the amount of data required is considerable, and the fact that the detailed destructive measurements required to provide these data are laborious. Due to natural diversity, sample size for developing allometric models should be rather large. Terrestrial laser scanning (TLS) has proved to be an efficient tool for collecting information on tree stems. Therefore, we investigated how TLS data for deriving stem volume information from single trees should be collected. The broader context of the study was to determine the feasibility of replacing destructive and laborious field measurements, which have been needed for development of empirical stem volume models, with TLS. The aim of the study was to investigate the effect of the TLS data captured at various distance (i.e. corresponding 25%, 50%, 75% and 100% of tree height) on the accuracy of the stem volume derived. In addition, we examined how multiple TLS point cloud data acquired at various distances improved the results. Analysis was carried out with two ways when multiple point clouds were used: individual tree attributes were derived from separate point clouds and the volume was estimated based on these separate values (multiple-scan A), and point clouds were georeferenced as a combined point cloud from which the stem volume was estimated (multiple-scan B). This permitted us to deal with the practical aspects of TLS data collection and data processing for development of stem volume equations in boreal forests. The results indicated that a scanning distance of approximately 25% of tree height would be optimal for stem volume estimation with TLS if a single scan was utilized in boreal forest conditions studied here and scanning resolution employed. Larger distances increased the uncertainty, especially when the scanning distance was greater than approximately 50% of tree height, because the number of successfully measured diameters from the TLS point cloud was not sufficient for estimating the stem volume. When two TLS point clouds were utilized, the accuracy of stem volume estimates was improved: RMSE decreased from 12.4% to 6.8%. When two point clouds were processed separately (i.e. tree attributes were derived from separate point clouds and then combined) more accurate results were obtained; smaller RMSE and relative error were achieved compared to processing point clouds together (i.e. tree attributes were derived from a combined point cloud). TLS data collection and processing for the optimal setup in this study required only one sixth of time that was necessary to obtain the field reference. These results helped to further our knowledge on TLS in estimating stem volume in boreal forests studied here and brought us one step closer in providing best practices how a phase-shift TLS can be utilized in collecting data when developing stem volume models.