3D object recognition is one of the most popular areas of study in computer vision. Many of the more recent algorithms focus on indoor point clouds, classifying 3D geometric objects, and segmenting outdoor 3D scenes. One of the challenges of the classification pipeline is finding adequate and accurate training data. Hence, this article seeks to evaluate the accuracy of a synthetically generated data set called SynthCity, tested on two mobile laser-scan data sets. Varying levels of noise were applied to the training data to reflect varying levels of noise in different scanners. The chosen deep-learning algorithm was Kernel Point Convolution, a convolutional neural network that uses kernel points in Euclidean space for convolution weights.

Throughout the past few decades, advancements in global navigation satellite systems (GNSS), such as the Global Positioning System (GPS), have resulted in real-time planimetric accuracy at the centimetre level, whether achieved using Real-Time Kinematic (RTK) or Real-Time Network (RTN) approaches. This study examines and characterizes the performance of RTK and RTN solutions at three test sites in Los Angeles County, California, in the United States of America. The solutions were characterized to examine the dispersion of vertical measurements in the context of different environments. The results from this study suggest that both methods have advantages and drawbacks; in particular, the data showed that the distribution and density of the network stations, cellular network coverage and environmental dynamics have significant impacts on RTN vertical precision, where the vertical precision for the RTN measurements were observed to be 2-4 times lower in comparison to the RTK measurements.

Abstract. Deep neural networks (DNNs) and convolutional neural networks (CNNs) have demonstrated greater robustness and accuracy in classifying two-dimensional images and three-dimensional point clouds compared to more traditional machine learning approaches. However, their main drawback is the need for large quantities of semantically labeled training data sets, which are often out of reach for those with resource constraints. In this study, we evaluated the use of simulated 3D point clouds for training a CNN learning algorithm to segment and classify 3D point clouds of real-world urban environments. The simulation involved collecting light detection and ranging (LiDAR) data using a simulated 16 channel laser scanner within the the CARLA (Car Learning to Act) autonomous vehicle gaming environment. We used this labeled data to train the Kernel Point Convolution (KPConv) and KPConv Segmentation Network for Point Clouds (KP-FCNN), which we tested on real-world LiDAR data from the NPM3D benchmark data set. Our results showed that high accuracy can be achieved using data collected in a simulator.

ABSTRACT The ability to accurately estimate the amount of stockpile material for construction projects can have a substantial impact on project budgets and schedules as well as public safety. Recent proliferation of small Unmanned Aircraft System (sUAS) platforms has made it possible to estimate these stockpiles using inexpensive aerial imagery and photogrammetry at a fraction of the cost and time compared to traditional methods. This study examines the quality of height and volume estimates from sUAS photogrammetry processed through Agisoft Metashape, Bentley ContextCapture, and PixElement, and compared with Global Navigation Satellite System (GNSS) based surveying and Terrestrial Laser Scanning (TLS) measurements. Measurements made on three co-located stockpiles occupying in total 345 m2 and a volume of 487.11 m3 showed that all three software produced results within 3% of TLS measured height and 2% of TLS volume overall, with greater local variation. Moreover, photogrammetry provided complete coverage while the TLS data had voids due to scanner obstructions. Considering that results could be further improved with refinements in technique, this study joins others in demonstrating that sUAS photogrammetry provides a viable alternative to conventional methods for estimating stockpiles.

Interest in small unmanned aircraft systems (sUAS) for topographic mapping has significantly grown in recent years, driven in part by technological advancements that have made it possible to survey small- to medium-sized areas quickly and at low cost using sUAS aerial photography and digital photogrammetry. Although this approach can produce dense point clouds of topographic measurements, they have not been tested extensively to provide insights on accuracy levels for topographic mapping. This case study examines the accuracy of a sUAS-derived point cloud of a parking lot located at the Citizens Bank Arena (CBA) in Ontario, California, by comparing it to ground control points (GCPs) measured using global navigation satellite system (GNSS) data corrected with real-time kinematic (RTK) and to data from a terrestrial laser scanning (TLS) survey. We intentionally chose a flat surface due to the prevalence of flat scenes in sUAS mapping and the challenges they pose for accurately deriving vertical measurements. When the GNSS-RTK survey was compared to the sUAS point cloud, the residuals were found to be on average 18 mm and −20 mm for the horizontal and vertical components. Furthermore, when the sUAS point cloud was compared to the TLS point cloud, the average difference observed in the vertical component was 2 mm with a standard deviation of 31 mm. These results indicate that sUAS imagery can produce point clouds comparable to traditional topographic mapping methods and support other studies showing that sUAS photogrammetry provides a cost-effective, safe, efficient, and accurate solution for topographic mapping.

This review paper explores at a high conceptual level cartography's potential role in the emerging field of indoor mapping. It introduces an interdisciplinary literature on foundational theories, approaches, and applications of indoor maps driven by advancements in indoor positioning systems and an accompanying desire to exploit those capabilities through maps. The review concludes that cartography, with its rich heritage in the mapping arts and sciences, can make important contributions as technologies, needs, and theories converge to make sophisticated indoor mapping a reality. This paper includes discussions of issues, challenges, and prospects for indoor mapping along with examples of possible new applications.

In recent years, growing public interest in three-dimensional technology has led to the emergence of affordable platforms that can capture 3D scenes for use in a wide range of consumer applications. These platforms are often widely available, inexpensive, and can potentially find dual use in taking measurements of indoor spaces for creating indoor maps. Their affordability, however, usually comes at the cost of reduced accuracy and precision, which becomes more apparent when these instruments are pushed to their limits to scan an entire room. The point cloud measurements they produce often exhibit systematic drift and random noise that can make performing comparisons with accurate data difficult, akin to trying to compare a fuzzy trapezoid to a perfect square with sharp edges. This paper outlines a process for assessing the accuracy and precision of these imperfect point clouds in the context of indoor mapping by integrating techniques such as the extended Gaussian image, iterative closest point registration, and histogram thresholding. A case study is provided at the end to demonstrate use of this process for evaluating the performance of the Scanse Sweep 3D, an ultra-low cost panoramic laser scanner.

One major challenge in creating indoor maps involves defining their levels of detail or LODs. While a consensus has emerged that indoor maps have at least two types of LODs, semantic and geometric, questions remain regarding their nature, their partitioning, and their relationships with each other as well as with other forms of LOD. Since semantics deals with the meanings of things, semantic LODs (SLODs) deal with the definition, classification, and partitioning of mapped entities. Unlike geometric LODs that are amenable to automation, SLODs have a more qualitative nature that defies automation and requires the careful application of human judgment. This paper proposes a framework for organizing semantic LODs by first classifying them based on the tangibility of mapped entities (i.e., intangible open spaces versus tangible physical features comprised of building structures and equipment and furnishings) and then partitioning each class based on the idea of permanence, defined here as an entity's tendency to remain stationary over time. A cartographic process for integrating SLODs with geometric and appearance LODs is also introduced along with several examples.

Indoor maps provide abstractions of the physical spaces where we spend most of our lives. General purpose indoor maps have historically taken the form of two-dimensional floor plans, commonly found in public venues such as shopping malls and cruise ships. Until recently, innovation and development of indoor maps have remained confined to urban planning and the building industry. Recent interest in indoor mapping for other applications has extended indoor mapping to 3D and to other domains, with a growing emphasis on commerce and general wayfinding. This paper reviews prevailing modeling standards and file formats relevant to the modeling and visualization of indoor spaces with the goal of assisting researchers and developers with finding appropriate formats for indoor modeling and visualization.

Low-cost digital photogrammetry using structure-from-motion (SfM) has made it possible for nearly anyone with a digital camera to create dense and precise point cloud models of the physical environment. However, the general requirement for large sets of photos in SfM can present a problem to those with limited resources or limited access to certain locations. This study examined the feasibility of using screen images of Google Earth’s proprietary high-resolution 3D models—not the crowdsourced SketchUp models—in creating point cloud and textured mesh models using SfM. The three study locations included a residential neighborhood in Tokyo, Japan; a portion of the University of California, Santa Barbara (UCSB) campus; and Mount Herard in Colorado. These locations represented a dense urban environment, a mixed environment, and a natural environment, respectively, where Google’s proprietary models existed. Light detection and ranging (LiDAR) data provided an additional data source for evaluating results at UCSB and Mount Herard. Simulated flights were “flown” in Google Earth at each location with screen capture software used to record 45° oblique video of the ground. Individual images were then extracted from the videos and used in Agisoft PhotoScan Professional, an SfM software program, to produce point cloud and textured mesh models of each location. Results of this study support the feasibility of using screen images of Google Earth for SfM modeling. While SfM succeeded in creating models for all three locations that visually resembled Google Earth’s own models, quantitative analysis showed that SfM worked best in the built-up areas of Tokyo and UCSB but struggled with the natural environment of Mount Herard. Comparison of sample distances within the SfM models and Google Earth showed planimetric errors of 1% or less and vertical errors of 5% or less for Tokyo and UCSB; however, absolute errors at Mount Herard—which was compared to LiDAR instead of Google Earth—spanned a range of under 10 m for areas of high relief to values exceeding 100 m for areas with low relief or low texture. The varying qualities of these models reflected not so much limitations of SfM but its reliance on a number of factors that impacted final model quality, such as image quality and operator skill in performing each step of the SfM workflow.