Terrestrial Laser Scanning PDF - Geomatics for Urban Analysis

GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Geomatics for Urban and Regional Analysis Applications of (Prof. Gabriele Bitelli)...

GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Geomatics for Urban and Regional Analysis Applications of (Prof. Gabriele Bitelli) Measurement terrestrial 3D scanning TLS (Terrestrial Laser Scanning) As-builts facilities and other 3D scanning methods Architecture Construction, BIM Roads, tunneling, mines Processing Archaeology Heritage documentation Restoration and preservation Geology, landslides Glaciology 3D Modelling Forensics, accident surveying Deformation monitoring Movies, games, animation Virtual reality … An expanding sector 3D terrestrial scanning A cloud of 3D points is generated, is an active system Differently from airborne Lidar, the coordinates of each point are defined in a device-centered 3D cartesian system, and the final results derive in general from the union of some clouds Georeferencing in an absolute system is not always needed The coordinates of the individual measured points are associated with intensity values (in the intensity map, highly reflecting points are displayed in a light grey pixel, highly absorbing points are displayed as a dark grey pixel, lack of a return is depicted as a black pixel) and (optionally) with color (RGB values, Red-Green-Blue) 1 t.o.f. Time-Of-Flight and Phase based Types of instruments 2 phase Terrestrial Laser Scanners 3 1 t.o.f. LASER (TLS) 4 2 Phase based Structured Light Projection (not laser!) GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 1 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Terrestrial laser scanners 1 2 for architectural applications Please note: unlike a total station, the instrument's rotation axis does not need to be positioned vertically. Another difference: the point on which the measurements are made (scan station) very often does not matter (= is not marked) Terrestrial laser scanners for architectural applications pointcloud The instrument, placed on a tripod, from a Terrestrial performs the simultaneous measurement of Laser Scanner slant range (distance) by a laser range (TLS) finder and the two associated angles by angular encoders in the horizontal and vertical planes passing through the centre of the instrument. The angular increments in both directions, comprising the azimuth and vertical rotations, can be set by the user. Typically the angular step sizes are set to identical values; thus the scanner provides an equal spatial sampling in an instrument-centered polar coordinate system The range can be measured using two main techniques: 𝑋 𝑑 sin 𝜔 cos 𝜔 1 - by Time-Of-Flight 𝑌 𝑑 sin 𝜔 sin 𝜔 - by phase measurement 2 𝑍 𝑑 𝑐𝑜𝑠𝜔 Selective measurement of significant points (e.g. total station) Two techniques for range (= distance) determination vs Non-selective measurement (scanner) (Kopáčik A., Erdélyi J., Kyrinovič P., 2020) GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 2 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli 1 Distance measurement by Time-Of-Flight (t.o.f.) 1 Pulse Ranging measurement Range calculation: pulses are emitted and Emitter and receiver of laser signal their travel time to and back from the Time measurement object is measured, multiplied by the speed of light c and divided by 2 range Mirror Beam deflection mechanism provides elevation and azimuth of the transmitted pulse Object Energy of the return pulse (intensity) and the optionally the color (RBG) is recorded Full waveform now recorded on some TLS instruments t.o.f. measurement requires very accurate time measurement 2 Distance measurement by a 2 Phase-based ranging measurement phase-based technique Another time-based measuring principle avoids using high precision clocks by modulating the power of the laser beam. The emitted light is modulated in amplitude and fired onto a surface. The scattered reflection is collected and a circuit measures the phase difference between the sent and received waveforms, hence a time delay. The distance is calculated by adding to the phase difference, Phase Ranging measured between the emitted wave and the received, the integer number of half wavelengths table not From the internal scanner reference system to updated an external one (object space) p The origin of the scanner space is the center of the instrument GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 3 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Choosing a ranging scanner instrument: Among the factors, choosing a ranging scanner instrument: Precision /Accuracy Acquisition speed Real maximum distance Wavelength of the signal Resolution and Divergence Field of view Ability to automatically recognize targets RGB acquisition, internal or external digital camera Data storage / download mode Easy to transport, easy to handle Power type Supplied software 3 3D Triangulators 3 Laser Triangulation for short-range applications and small objects Laser emitter + digital sensor Triangulation exploits the cosine law by constructing a triangle (camera) are at a using an illumination direction (angle) aimed at a reflective surface known (small) and an observation direction (angle) at a known distance (base relative distance distance or baseline) from the illumination source Object Laser Image on CCD sensor (x,y) Base (known) Unknown point Camera The laser is emitting a point or a line (or multiple lines), normally in red (visible) Laser triangulation vs Stereoscopic photogrammetry 3 Laser triangulation P Py Px An illustration of (a) triangulation and (b) stereo-based 3D acquisition systems. A The angle  is known (mechanical rotation of the mirror) triangulation-based system is composed of a light projector and a camera, while a The angle  is calculated once the image point corresponding to P stereo-based systems uses two cameras. In both cases, the depth of a point on the (and therefore its coordinates) has been recognized on the image 3D object is inferred by triangulation. sensor (the focal lenght f is a known value) GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 4 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Small objects can be placed on a swivel base and 3 rotated (8-12 acquisition, with a rotation 30 or 45 degrees between two successive) From the exposition KINKU 4 Structured light projection scanners 4 Structured light projection scanners They use the principle of triangulation but do not use the laser. They emit a white or blue light on which a pattern is imprinted (e.g. a regular grid of points, or parallel bands with variable frequency) which deforms in accordance with the geometry of the surface of the scanned object, and which is decoded. The image is continuously acquired by a video camera, which records several frames per second with a certain frequency. Pattern projector Small distances and narrow field of action High scanning speed Object profile High accuracy (submillimeter) High density Structured light projection scanners 4 Structured light projection scanners They can have a fixed or pivoting structure (= they can be moved freely by hand) GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 5 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli 3D precision: 0.1 mm 3D precision: 50 microns Archaeology: surveying a tablet with cuneiform alphabet 3D reproduction for museum replicas Dark clay tablet - EG4013  Dark colour of the The application of the A further improvement to the material algorithm provides good engraved text visualization can be  Small dimension of the results in terms of provided by applying artificial lights wedges (∼ 3 mm) to the digital model enhancement of the wedges Before (3D textured model) After (3D filtered model) Varying the direction of the beams 200000 poly faces of light, it is possible to enlighten 1000000 poly vertices certain parts of the text of particular interest F6 Volumetric structured light projection scanner F6 Volumetric structured light projection scanner Stonex F6 Smart (by Mantis Vision) is a structured light triangulating portable scanner that works in the near infrared field. F6 smart uses the structured light 3D scanning technique, projecting a pattern of light on the object to be detected. A Near Infrared sensor (NIR, 850 nm) examines the geometry of the model and calculates the distance to each point in the field of view. Operating with infrared, it is independent of ambient lighting. Operates for distances from 60 cm to 4 m, acquiring 8 fps (frames per second) with accuracy 0.2 - 0.1% of the distance (4 mm at 4 m) GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 6 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Characteristics External camera support IR Camera (NIR sensor) RGB camera The system acquires the 3D data by projecting infrared light through a m ask thatgenerates a proprietary pattern. This projected infrared lightis reflected offthe surface and captured by both the RGB cam era Projector and the N IR depth cam era. Applications System / Use / Accuracy and range Architecture, as-built documentation Plant engineering Automotive Cultural heritage Forensic documentation Virtual / Augmented Reality... Design/Planning of the survey Focusing on terrestrial laser scanners for architectural Pointclouds acquisition Topographical survey purposes, survey of structures and infrastructures… (scanstations) of the targets (optional) Editing and filtering of the pointclouds Alignment / registration on a single reference system Quality Control Fusion Workflow Polygonal mesh generation Mesh edit (data cleaning, hole filling,...) Mesh optimization, Decimation, etc. For a comparison of current commercial TLS: https://geo-matching.com/terrestrial-laser-scanners Final product generation (point cloud, rendered mesh, 2D / 3D sections, etc.) GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 7 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli The laser scanning equipment t.o.f. vs phase-based Phase difference instruments generally offer slightly lower accuracies than time-of-flight instruments, for which the error is not much affected by distance. However, they can be considered more advantageous if other characteristics are considered, such as the scanning speed (higher than that of time-of-flight scanners) and often the cost. Low precision High precision Distance The laser and the risk classes for exposure to eyes The laser and the risk classes for exposure to eyes Class1 Class3B Class 1: safe for occasional exposure over 7 m Class 3B: safe for occasional exposure over 160 m Riegl VZ400 1 – laser beam 2 – polygonal mirror 3 – optical head 4 – display and keypad 5 – TCP/IP interface 6 – additional TCP/IP Ethernet interface 7 – wireless LAN antenna 8 – USB storage device 9 – camera 10 – laptop 11 – mobile device 12 – operating software GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 8 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Leica RTC 360 High-speed 3D laser scanner with integrated HDR spherical imaging system and Visual Inertial System (VIS) for real time registration: automatically pre- register point cloud data in the field to quickly conduct quality control checks and improve productivity With a measuring rate of up to 2 million points per second and advanced HDR imaging system, the creation of coloured 3D point clouds can be completed in under two minutes. Small and lightweight Pointclouds in a device-centered reference system From the point clouds (hundreds of thousands, millions of points, etc.) it is possible to obtain the 3D model of the object and the derived products (profiles, contour lines, etc.) A fundamental premise about the reference systems… but initially the points derived from each scan have coordinates in a local system with (0,0,0) in the instrumental center The Which coordinate systems? reference SOCS – Scanner Own Coordinate System each scan position has its own system system problem PRCS – PRoject Coordinate System local system of the entire project GLCS – GLobal Coordinate System If the same point is surveyed from two different scanstations absolute system (UTM, WGS84, national geodetic-cartographic system) (even if it is very difficult to be exactly the same point…) it will have two different coordinate triplets The point clouds obtained from two different stations must be aligned / registered with each other to have a unique representation of the object. GLCS GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 9 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Design and Data Acquisition process Design and Data Acquisition Operations: 1) Design of the survey, depending on the purpose and the situation 2) Establishing the measuring scheme, position of the scan-stations (front, center-right-left,...) 3) [when needed…] Setup of targets and topographic survey 4) Scan Resolution Setting 5) Acquisition of the single scans Acquisition parameters different for each case, depending on: - Object to be surveyed (size, shape, specific geometrical characteristics...) - Logistic constraints (area available for the scan stations, need of acquistion from high positions, …) - Scale of the survey > precision required - Instrumentation available (scanner and other geomatics devices…) - Aim of the survey On the field Design of the survey Take a walk around the field site before setting anything up. Identify scan positions and target Material and geometrical aspects of the object must be considered positions (if you want to use targets for alignment together with features of the TLS and reference system definition). Choice of the acquisition distance, also based on on-site If possible use more than 4 targets (the more the better) in the overlap area to register between constraints, to achieve the desired point density them two scan positions. Overlap between the single scans: 30-40% for a subsequent reliable Set up targets. Draw a sketch with target alignment between point clouds arrangement and names If needed, survey them by total station (if the object is complex, you need a lot of targets NOTE on the relationship between survey precision and object shape: if the and a topographical three-dimensional object is highly three-dimensional, with asymmetric irregularities, the semi- network to calculate their position in a unique automatic alignment algorithms can well lock each scan into each other, but reference system) if the object has a reduced three-dimensionality the measuring uncertainty Scan Position 1 360-deg panorama scan + image acquisition if must be minimum (choice of instrument, distance, etc.), because the desired algorithm would tend to align along the ripples in the surface that are (Target finescan) produced by the instrumental errors Area of interest finescan Scan Position 2 and beyond… Same as Scan Position 1 (for the target finescan use corresponding points with previous scans) Scan Example acquisition schemes GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 10 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Problems related to the geometry of the object Partitioning a scan In some cases, where the distance from the area of interest varies considerably, it may be useful to partition the scan by imposing different steps for the different areas to ensure an almost homogeneous density on the entire object Influence of surface discontinuities on Possible occlusions with respect to the If you impose the same angular step for the scan r1 that guarantees the the scan (Lato et al., 2010) inclination of the laser beam required density on r3, the scan r1 will take a long time and will produce a (Sturzenegger & Stead, 2009) very heavy point cloud to handle Summarising…(I) Summarising… (II) No matter which scanner is used, the basic principle is the It is very important that the right point same. A laser is fired out and for every surface that the laser hits density is determined prior to commencing a point in space is recorded (xyz). At the same time the scanner will record the reflectivity of the surface giving an intensity value, a project. When reducing the point density, and nowadays most scanners also have in built or connected key features, edges etc. may not be visible cameras which provide an RGB (colour) value to each point. in the results. The points are captured at speeds of up to 1 million points of To join multiple scans together different data per second, creating a very dense point cloud of data. solutions can be adopted. This can be done by using either targets in All laser scanners work via line of sight → on a typical project multiple scans the scans (whose coordinates can be need to be taken from different positions to ensure a complete data set. unknown or known through a topographical survey, e.g. by a total station) or by When scanning large objects such as buildings we can allowing enough overlap in the scans to scan for example at a point density of 3mm at 10m. This means there is a point spacing of 3 mm between each perform a semi-automatic or automatic point at a distance of 10 m from the scanner. As you cloud registration (alignment) by move further away this spacing increases, so at 20 m recognizing common features (cloud to the distance between each point will have increased to cloud). 6mm. New technologies support today a pre- registration on the field. Scan time Divergence The scan time (for laser systems) or acquisition time (projection systems) ranges from a few seconds to several minutes Time taken for a 360 ° scan (Riegl VZ2000i Frequency 1200 KHz) With higher laser beam divergence values, the spot area is greater and the content is more mediated: loss of detail GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 11 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Maximum measurement range as function of target material refelectivity (scanner Riegl VZ400, laser wavelength: near infrared) Distance laser – object, stepping angle and point density With the same angular resolution, the density of the points detected on the object Density related to object features varies with the distance and with the inclination of the surface to be detected with respect to the instrument Case: Architecture – Cultural Heritage it can also be thought of in relation to the final scale of the resulting representation or the expected final products Footprint size Some considerations about example: Scanner OPTECH ILRIS 3D resolution / density Does not make sense that the density is greater than the instrumental precision (do not set-up a distance among points less than 5 mm if the instrument provides the coordinates of the points within ± 5 mm) Density must be related to the nominal scale of the final product (e.g. for 1:20 scale: +/- 4.0 mm) Regions of overlap between scans are often large, which may result in much higher point spacing when point clouds are merged. The denser the point cloud, the more noticeable will be the random noise, especially when overlapping data are combined Instrument manufacturers report the maximum point spacing (or angular resolution) as being higher than the width of the laser beam itself. Choosing too small the angular spacing can therefore result in “blurring” of the data caused by oversampling the surface. It is best to ensure that the resolution of the collected data is the same in the horizontal and vertical direction. Unless there is a good reason for having different values for each axis, this can complicate processing, and even result in lower data quality if resampling is required. GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 12 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Footprint given the point spacing Scanning resolution (density) vs Error Mesh generated from a point cloud obtained by surveying, with a scanner OPTECH ILRIS 3D density of 1 mm, a flat surface by means of a scanner that has an error of 2 mm. Scanning with a ratio density/error equal to 1:2 or 1:3 can lead to a mesh with excessive angles between adjacent faces. The same surface generated from a scan with a point density of 10 mm. It must be furthermore considered the size of the footprint, for example 4 mm in diameter. The distance measured is the average of all distances for the area illuminated by the laser. If we fix a scan resolution of 1 mm, and then move for 1 mm a spot with a diameter of 4 mm, we are re-acquiring an information that for 3/4 is equal to that of the previous spot, and therefore is mainly influenced by the noise linked to the instrumental precision. Interaction signal – object geometry apparent spot position caso ideale effective spot position Depending on the geometry of the object and the angle of incidence of the signal several cases may occur Effects due to object edges Data voids (gaps on the acquisition process) RGB image acquisition The joint acquisition of the RGB data can help to better understand the object (eg. in this case in which the geometry was only partially detected) and as aid for the collimation of points / features during alignment In some cases, lighting should be checked (shadowing in outdoor, lower light sources in indoors that produce reflections or excessive contrast reductions, etc) There is normally no white calibration (WB) for each instrument position, and so you can have different hue for each scan position when texturing the 3D model GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 13 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Point cloud RGB acquisition options visualization coaxial with laser beam by a reflex camera coupled with laser scanner Intensity vs RGB Camera mounting on a Riegl VZ-400 Visualization of the primary data in the form of maps of distance / intensity / (RGB) / point clouds Data formats for point cloud target with high reflectance at the signal wavelenght  maximum value in the recorded intensity intensity azimuth and zenith 3D angles + coordinates distance in the local from the instrumental instrumental system distance center RGB Editing and filtering of the pointloud Point cloud cleaning Removal of disturbing elements or elements unrelated to the geometry of the object of interest (scaffolding, railings, vegetation, people,...). Editing of data affected by errors incompatible with the characteristics of the survey and the physical model. The points that have a high probability of not belonging to the object surface are deleted It can be performed manually or in some cases entrusted to automatic procedures (e.g. selection based on a filter on min / max distances) (GECO Lab.) GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 14 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Filtering / noise reduction Point cloud artefacts In theory, data should be unfiltered, because it means that the choice of instrument was not consistent with the physical and material characteristics of the object, but very often we do not have a choice between several instruments. Use of robust filters (e.g. median filter). Different forms of point cloud artefacts, shown here in the case of a curve shown in 2D Point decimation (data reduction) Alignment and fusion of the pointclouds To reduce the spatial density of the points (resampling) and/or to make the cloud more homogeneous and consistent in respect to the nominal scale of the survey. Often the resulting points are at the same time disposed on a regular grid (using an octree). Aligning / registering the point clouds Approaches for alignment / registration In the case of two 3D-laser scans, six degrees of of pointclouds freedom need to be solved since shift (3 translation/shift parameters, t) and rotation (3 angular parameters, R) can be applied to the three cardinal axes X Y Z so that the coordinates 1) Registration with use of targets of the point cloud B are expressed in the reference system of A. 2) Semi-automatic registration by ICP methods 3) Pre-alignment by using other sensors (e.g. by Computer Vision techniques) A B … some preliminary considerations Before alignment After alignment GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 15 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Aligning / registering the point clouds Alignment of cloud points Approach 1: Registration with use of targets Recognizing some common points (belonging to overlapping areas) on two clouds acquired by different scan stations it is possible to 1) Scanning of targets visible in the scene with a high scanning step (they could be known in an object or absolute reference system) calculate the registration parameters (3 translations + 3 rotation angles) to move each cloud towards the other, or towards a common 2) Extraction and position calculation for the center of the targets final reference system (it could be for example the system of the first 3) Identification of at least 3 common targets among 2 neighbouring scans scan or a system defined by a topographic survey, ensuring in this on the overlap area way also the verticality of Z axis) 4) Assignment of the same nomenclature to the target group for each of the 2 range scans Optional: if the coordinates of the targets are known (in an absolute An often overlooked effect of reference system or a system related to the object) they can be entered registration is that it can: in this phase a) create errors that easily 5) Co-registration of the 2 scans by the calculation of the 6 parameters of overcome the error of the the geometrical transformation If I measure the distance between two points on a scanner itself map with a short ruler, the final error will not 6) Repeat the process between groups of 2 or more scans with 3 or more b) cause errors that affect all depend only on the inaccuracy of the ruler itself common targets. but on how it was subsequently placed by other linked scans aligning it along the distance to be measured: error propagation 7) Creating a set of scans aligned with each other. 8) Evaluation of the quality of the process Use of targets as reference Targets recognition and link between 2 scans Targets: two or three dimensional diffusive or retro-reflective objects that serve as reference points for scans (some targets must be common between scan positions, the more the better). They can be collimated by surveying instruments (total stations) to insert the final cloud point into a specific object reference system; in some environmental applications (outdoors) the coordinates can be obtained by GNSS. The center of the target (e.g. a circle or a sphere) is the effective point to be considered in registering scan positions Used for instance in architecture, especially with high resolution TOF scanners that automatically detect the targets and perform a high resolution scan on there High reflectance target, with very dense Automatic recognition and Target georeferenced by a high resolution scanning coincident GNSS antenna scan data, automatically detected on the of a circular target intensity map (RiScan Pro by Riegl) 2D targets Circular reflective Checkerboard Examples of 2D targets: on tripod, double-sided, wall-mounted GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 16 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Cilindrical targets Spherical targets the reference is the centre of the sphere, which is identified in the same way by any position in space Alignment of cloud points Approach 2: Alignment by (semi-)automatic procedures (ICP) It can be difficult, or can require a lot of time, to place and survey targets at the scene: algorithms have been developed that can take advantage of the shape of the object as a reference to be to matched by the different scans, then exploiting the data redundancy They are normally semi-automatic, requiring an initial expeditive relative positioning of the single clouds, and have largely improved over the last years Note: if you need to scan a building with a lot of rooms requiring separate scans, it is better to use targets (error propagation!) Alignment of two pointclouds Alignment of two pointclouds by (semi-)automatic procedures by (semi-)automatic procedures Solving for a rigid-body transform from the results of two First step: pre-alignment, by: scans – manual expeditious identification of corresponding points in the overlapping area, also with the help of RGB Rigid registration problem > – manual relative arrangement of the pointclouds solving for a rigid-body – measurement of the individual location and orientation of two scans transform (i.e., translation and by additional sensors (eg by Computer Vision approaches) which rotation, or 6 total degrees of today may be frequently found in up-to-date laser scanners freedom in 3D) that minimizes the distance between two Second step: correspondence search between the two sets of points, partially-overlapping meshes. and computation and application of 6 parameters transformation using We’ll focus on iterative iterative algorithm until the average distance is minimum → algorithms that converge to a minimization of an objective function whose value is recalculated local minimum, so we’ll assume varying the relative orientation between the clouds that we have an initial guess already. This approach is in general named: ITERATIVE CLOSEST POINT (ICP) GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 17 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli ICP Aligning 3D Data In a first phase a expeditious alignment is performed manually (moving the 2nd cloud up to overlap the 1st in the common area, or by selecting approximately 3 or more common points) Depending on the implementation, correspondences are either determined by the shortest distances between one point to another or from a point to a plane. Based on this information registration parameters are computed and applied to one of the datasets (the yellow boxes). The black dotted arrows between the orange and yellow boxes indicate that these steps are iteratively repeated until a convergence criterion is fulfilled, and the After a minimization algorithm, the first solution is still wrong but if you final solution has been found → different correspondences are established started out reasonably close, this process got you closer… during the course of the algorithm. Aligning 3D Data Search of corresponding points … and iterate to find alignment, you can repeat the process and converge to the right answer The Iterative Closest Points (ICP) algorithm converges if Point-to-Point approach starting position is “close enough“ Point-to-Plane approach Note: it is important that the overlapping area is large, otherwise there is the risk of finding only a “local minimum” in the distance between the two clouds Point-to-Point approach Point-to-Plane approach The objective function in this case is the distance between the spatial For each point of the mobile cloud are searched those position of a point in a dataset and the plane tangent to the belonging to a sphere of radius double the average distance corresponding point in the other. In total will be the sum of the between the points, and the closest is chosen. squares of the distances... Association between point elements and two-dimensional elements. The operation is repeated for all the points of the mobile cloud. Increased speed of the convergence process (an order of magnitude lower than the point-to-point), better automatic recognition because the In the minimizing phase there is the risk to consider also erroneous false points create fewer problems. points, which can cause many iterations and sometimes the problem For certain types of surfaces (overlapping area almost flat or nearly of skidding/shift between the individual range maps. uniform curvature) the point-to-point approach is instead preferable. GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 18 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Summarizing the ICP operation (I) Summarizing the ICP operation (II) The number of scans acquired to cover the entire surface of the Some variations of ICP can also work with multiple scans object is related to its geometric complexity simultaneously and not just two, using e.g. the information resulting from the closure of surfaces, when possible. If the object's structure is substantially on one or two main The single clouds must be aligned in the same reference system dimensions (e.g. a car door) it is more difficult to eliminate the and then merged together to make a final unique 3D model propagation of the alignment error. In these cases is better to have known coordinates of artificial targets. A cloud is normally used as the initial reference and by a semi- In general, the lack of particular geometrical features on the automatic procedure is found the optimal positioning of a second object can generate problems of convergence and global with respect to it, and so on. In this first phase, the choice of the alignment errors (e.g. industrial applications) homologous points usually takes place with a first manual positioning, aiming at providing a starting point for the iterative procedure. Problems of propagation of co-registration Risks for ICP-based quality assurance errors using ICP approach ICP always finds a solution – yet not necessarily the right one. Hence, visual inspection is always recommended when using this algorithm. However, simply looking at data is quite subjective. Hence, a more sophisticated solution for sound quality assurance is needed → external control by an independent Using ICP can increase the method of measurement risk of propagating the registration error that original position in the Example Two entirely different occurs between pairs of space of the 6 scans datasets are registered with a commercial solution. The sample successive scans. size indicates for how many points the ICP should try to find correspondences. Leaving this Examples: high probability value unrestricted would be very computationally demanding. The of accumulating error on a second value determines the long linear job such as a largest distance between two road and dealing with a points from two datasets that can very regular man-made form a correspondence. In this case the produced result is (Wujanz et al. 2016) structures such as the residual misalignment non-sense: a very limited average inside of a tunnel. between S1 and S6 error is not always rappresentative of a good result Alignment of cloud points Approach 2: Latest operational solutions to perform a pre- alignment during the data acquisition phase and optimize the scanning phase Alternative ways to register the point clouds together during the survey itself, with subsequent optimization afterwards Identification of targets by photographic means (Computer Vision algorithms) Use of HDR images Documentation with multimedia elements connected to the 3D model Presentation and sharing of models Equipment examined: time-of-flight laser scanner Leica RTC360 GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 19 G. Bitelli GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli The instrument is capable of collecting spherical images in High-Dynamic Range (HDR). HDR is a popular mode in photography: the same image is collected with different exposure settings to get the best exposure value for each pixel. VIS Visual Inertial System to make the alignment / registration operation between two scans easier from the moment of the acquisition itself, it calculates the 6 roto-translation parameters in real time. is a SLAM (Simultaneous Localization and Mapping) type solution of visual type, i.e. based mainly on image processing algorithms operating on the images acquired by the 5 cameras (four on the corners of the scanner and one towards the nadir) 2nd position unknown: 2nd position determined in misaligned clouds real time: aligned clouds GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 20 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli 5 cameras INS SLAM algorithm continuously and alternatingly determines: A new pose, i.e. the six degrees of freedom of the laser scanner determined by a spatial resection based on recognized previous 3D points (localization) Including the new pose, determining further 3D points through a forward intersection and expanding the map accordingly (map creation) At the beginning, the map is initialised by selected 3D points from the first scan. For these points, optically significant features with high contrast in the image such as corners and edges are selected. Their change in position in the camera images during the movement of the scanner can be determined robustly using a feature tracking method. Feature extraction dalle immagini delle 5 camere Recap: App Cyclone FIELD 360 We have seen 3 approaches for alignment / registration of pointclouds… 1) Registration with use of targets 2) Semi-automatic registration by ICP methods 3) Pre-alignment by using other sensors (eg by Computer Vision techniques) GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 21 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli The final cloud: Fusion process Polygonal mesh generation Once all the 3D clouds are aligned and edited, they can be merged into and processing a single cloud with an automatic operation (fusion), and a single final mesh produced All the redundant information is lost and you lose track of the initial data Necessary to verify that the model obtained retains the geometrical and morphological characteristics of the original, minimizing topological errors and gaps Aligned scans associated Fusion carried out Final 3D model with different colors From the point cloud to a surface (the mesh) Meshing procedure: Delauney triangulation The triangulation can be realized with the Delauney method, building the triangles connecting the points. A Delaunay triangulation maximizes the minimum angle of all the angles of the triangles in the triangulation; they tend to avoid sliver triangles. The Delauney criterion provides that three points form the vertices of a triangle if the circle circumscribed with them contains no other known points. NO Other points are contained within the circumcircle Mesh generation and optimization Subdivision and optimal redistribution of polygons (Remesh). Regularization or in some cases increase to better describe the surface Reducing the number of polygons: it must be made according to the A Delaunay purpose of the product to be obtained triangulation in the plane with circumcircles shown 36000 polygons 20000 polygons (NRC, J.C. Beraldin) GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 22 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli (GECO Lab.) Multi-resolution mesh Thinning and remeshing Mesh Editing The software packages present different solutions for the correction (editing) of the meshes, in manual, semi- automatic or automatic way There may be topological errors, gaps and noisy areas The detection / correction of topological errors differs from software to software Hole filling procedures Topological Gaps: due to shadows or reflection problems errors On planar areas you can use fast inspection techniques On complex areas: fitting surfaces Typologies of NB In any case it is good to clean up the edges of the gaps preliminarily topological errors automatically detected by Polyworks (above) and Rapidform (below) software Degenerate triangles: 2 or 3 vertices of a triangular mesh face are equal: the triangle degenerates into a point or segment. Duplicate triangles: two triangles of the same mesh have the same vertices Degenerate edge: edge shared by more than two triangles (= Non-Manifold face) Inconsistent edges: an inconsistent edge is shared by two adjacent triangles (a)gap in a flatarea;(b)gap in a curved part;(c)large gap w ith the with opposite orientation (normal inverted) (= Reversed/Unstable Face) construction of"bridges"to reduce the gap area into sm allerportions Crossing Face: faces intersecting with the mesh without connecting to any vertex (Guidi, Russo, Beraldin, 2010) Redundant Face: abnormal faces that can be found in a mesh when in a vertex converge a different number of vertices and edges. GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 23 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Final products generation Timing for the development of a model It depends on many parameters (environmental, geometry object, sampling in acquisition, operator, instrument type, expected products,...) The point sampling defined in acquisition influences greatly the whole process: the change of an order of magnitude can lead to an increase of 10 times the time of the overall process Object: 20 m x 20 m man hours Processes and Timing Embedding interactive 3D object in a.pdf file Cultural Heritage Design A 3D object can be also inserted inside a.pdf file (U3D format) to be examined interactively (Guidi, Russo, Beraldin, 2010) 3DS - 3D Studio BLEN - BLENDER File formats DAE - COLLADA DXF - AutoCAD for the final FBX - Autodesk exchange products of geoTIFF glTF Notes about some common formats: 3D modeling LWO - Lightwave OBJ.PLY OFF PLY.STL PTS PTX.OBJ Many formats, some proprietary and SC1 - Sculptris some open (neutral): e.g. STL, OBJ, SCL - Pro/engineer FBX, COLLADA etc. SKP - Google sketchup STL Widely used in 3D-printing, video games, TRI movies, Architecture, Medicine, V3D Engineering, Earth Sciences. WRL - VRML X3D Each industry has its own popular file X3DV formats for historical and practical XSI - SoftImage ZTL - Zbrush reasons. XYZ Problems for interoperability… … GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 24 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli.PLY Polygon File Format (Stanford Triangle Format).STL Especially designed for 3D scanners STL (STereoLithography) is one of the most important neutral 3D file formats in the domain of It manages the object list of flat polygons, to which attributes 3D printing, rapid prototyping, and computer aided can be associated such as: color and transparency, normal manufacturing. to the surface, texture and data confidence values. The front It is native to the stereolithography CAD software and back of a polygon may have different properties. made by 3D Systems. Format: ASCII or binary STL is one of the oldest 3D file formats and was created in 1987 by Chuck Hull. He also invented the world’s first stereolithographic 3D printer. The STL file format was created subsequently as a simple way of transferring information about 3D CAD models to this 3D printer. STL encodes the surface geometry of a 3D model approximately using a A normal to a surface triangular mesh. It ignores appearance, scene, and animations. It is one of at a point is the same the simplest and leanest 3D file formats available today. as a normal to the The STL format specifies both ASCII and binary representations. Binary tangent plane to that files are more common since they are more compact. surface at that point..OBJ OBJ is a geometry definition format developed by Wavefront Technologies for its animation software. Drainage control structure for The format is open and has been adopted by other manufacturers; is a canal: from the TLS survey universally accepted. to the 3D model ASCII and binary (.MOD) version The OBJ file format supports both approximate and precise encoding of surface geometry. When using the approximate encoding, it doesn’t restrict the surface mesh to triangular facets. If the user wants, he can use polygons like quadrilaterals. When using precise encoding, it uses smooth curves and surfaces such as NURBS. The format support also the texture definition The OBJ file format stores the position of each vertex, the UV position of each texture coordinate vertex, normals, and the faces that make each polygon defined as a list of vertices, and texture vertices. Vertices are stored in a counter-clockwise order by default, making explicit declaration of normals unnecessary. Dagostino & Wood, Inc. Surveys repeated over time for dynamic phenomena Monitoring of a progressive collapse phenomenon on a wall in an open pit mine Notre-Dame: la numérisation de la cathédrale en 3D e.g. monitoring of degradation GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) 25 G. Bitelli  GEOMATICS FOR URBAN AND REGIONAL ANALYSIS (2024/2025) G. Bitelli Applications (I) Applications (II) Architectural surveys: Environmental surveys and exteriors monitoring: landslides Environmental surveys and monitoring: quarries Architecturals surveys: and open cast mines interiors Cultural Heritage, Roads design and Archaeology management Tunneling Applications (III)