UAV/LiDAR-system har utvecklats snabbt under de senaste åren. De har blivit kompaktare och kan generera en hög punkttäthet med låg mätosäkerhet. Samtidigt har prisnivåerna sjunkit, vilket medfört att ett UAV/LiDAR-system nu kan införskaffas för några hundra tusen kronor. Det finns även mer avancerade UAV/LiDAR-system för ett par miljoner kronor.  Syftet med denna studie är att undersöka vilka skillnader som finns i kvalitet mellan två UAV/LiDAR-system i olika prisklasser. Framför allt kommer osäkerheter vid skapande av digitala terrängmodeller att studeras. De system som ingår i studien är drönaren DJI Matrice 300 RTK med LiDAR-sensorn DJI Zenmuse L1, och drönaren Microdrones md4-3000 kombinerad med LiDAR-modulen RIEGL miniVUX-1DL (mdLiDAR3000DL).  Undersökningen utfördes på en grustäkt i Rörberg utanför Gävle. En fri stationsetablering gjordes för att kunna mäta in stöd- och kontrollpunkter samt fem kontrollprofiler i terrängen. En flygning per drönare genomfördes över området. De erhållna punktmolnen bearbetades och jämfördes sedan mot kontrollpunkterna i höjdled. Även mängden brus i punktmolnen studerades. Därefter skapades digitala terrängmodeller som jämfördes mot kontrollprofilerna enligt metoden i SIS-TS 21144:2016.  Innan georeferering mot stödpunkter fanns det relativt stora skillnader mellan respektive punktmoln gällande avvikelsen i höjd. För DJI-punktmolnet blev till exempel medelavvikelsen 49 mm, medan mdLiDAR3000DL genererade en medelavvikelse på –21 mm. Efter brusreducering och inpassning av punktmolnen blev resultaten mer lika. Till exempel hade de två systemens punktmoln 13 mm i både standardavvikelse och RMS. Även de uppskattade brusmängderna var relativt lika. Den enda märkbara skillnaden var att mdLiDAR3000DL gav fler låga punkter procentuellt sett. Kontrollen av respektive DTM gav också ett jämnt resultat. DJI-systemets DTM hade en variationsvidd på 43 mm, medelavvikelsen 0 mm och standardavvikelsen 11 mm. MdLiDAR3000DL:s DTM hade 42 mm i variationsvidd, 0 mm i medelavvikelse och 8 mm i standardavvikelse.  Slutsatsen som kan dras är att det främst finns skillnader i osäkerhet innan någon databearbetning har genomförts. När punktmolnen brusreduceras och georefereras jämnas resultaten ut. Hanteringen av punktmoln har följaktligen en avgörande betydelse för osäkerheten i slutändan. Båda UAV/LiDAR- systemen kan därmed generera högkvalitativa slutprodukter. / UAV/LiDAR systems have developed rapidly in recent years. They are now more compact and can generate high point density with low uncertainty. The price level has also decreased, which entails that UAV/LiDAR systems are nowadays available for a few hundred thousand SEK. At the same time, more advanced UAV/LiDAR systems exist that cost a couple of million SEK.  The aim of this study is to investigate quality differences between two UAV/LiDAR systems in different price ranges. The focus will be on studying uncertainties when creating digital terrain models. The systems in question are the drone DJI Matrice 300 RTK with LiDAR sensor DJI Zenmuse L1, and the drone Microdrones md4-3000 combined with LiDAR sensor RIEGL miniVUX-1DL (mdLiDAR3000DL).  The study was made in a gravel pit in Rörberg outside of Gävle. A free station was set up to be able to measure ground control points (GCPs), check points, and five profiles in the terrain. One flight per drone was conducted over the area. The obtained point clouds were processed and later compared to the check points with respect to height deviations. Also, the noise level in the point clouds were studied. Digital terrain models were created and compared to the profiles according to the method in SIS-TS 21144:2016.  Before georeferencing against GCPs, there were relatively large differences between each point cloud in the height component. The DJI point cloud had for instance a mean deviation of 49 mm, while mdLiDAR3000DL generated a mean deviation of –21 mm. After noise reduction and fitting of the point clouds, the results were more even. For example, the two points clouds resulted in 13 mm in both standard deviation and RMS. Also, the estimated noise levels were rather similar. The only noticeable difference was that mdLiDAR3000DL generated a larger percentage of low points. The check of each DTM also resulted in similar results. The DTM of the DJI system reached a range of 43 mm, a mean deviation of 0 mm, and a standard deviation of 11 mm. The DTM of mdLiDAR3000DL obtained a 42 mm range, 0 mm mean deviation, and 8 mm standard deviation.  Mainly, there are differences in uncertainty before any data processing is carried out. When the point clouds are noise reduced and georeferenced, the results even out to a large extent. The handling of point clouds thus has decisive importance for the final uncertainty. Thus, both UAV/LiDAR systems can generate high-quality products.

Med hjälp av luftburen laserskanning kan stora mängder av LAS-data samlas in. Lantmäteriet, en statligmyndighet vars uppgift är att kartlägga Sverige, har gjort en rikstäckande insamling av LAS-data. Denna data ärtillgänglig för allmänheten och kan hämtas gratis från Lantmäteriet. Då insamlingen är rikstäckande ärupplösningen på LAS-datan lägre än vid en lokal insamling. Punkttätheten på LAS-datan ligger på 1–2 punkterper kvadratmeter.Detta arbete gjordes på uppdrag från Arvika kommun då de ville veta om det är möjligt att extrahera och karteraArvika tätorts urbana växlighet från Lantmäteriets gratis LAS-data. Det som efterfrågades från kommunen var ettpunktlager som visar ungefärliga positioner på träd över 3–5 meter. Punktlagret skulle ha attribut medinformation om trädens storlek. Detta hade varit användbart både för kommunens GIS-avdelning för attsimulera skuggning vid planeringar, modellera träd och framtagning av informativ växlighetsstatistik. Det skulleäven vara användbart för parkavdelningen som tidigare haft brist på data över deras urbana växlighet.Denna uppgift genomfördes genom att kombinera olika funktioner i ArcGIS Pro. Först konverterades LASdatasetet om till raster. Detta raster tas sedan multiplicerat med -1 och därmed vänds. Detta inverterade rasteranvänds i hydrologiska analysfunktioner som baserar sig på att jämföra celler med dess grannceller. På detta vistas ett punktlager fram som visar lokala höjdtoppar. Detta lager matas med höjd från raster och även ett NDVIvärde. Detta NDVI-värde tas fram från två uttagna band (band 1 och 4) från Lantmäteriets ortofoto.Ett diameter-värde beräknas genom att avgöra antalet celler som trädets krona består av. Detta värde anses bliväldigt generaliserat, men har en relativt konstant differens och bör därför kunna användas för storleksindelningav träden. De framtagna lokala höjdpunkterna filtreras även med ett par kriterier som de måste uppfylla. Dessakriterier baserar sig på några av de informativa data som framtagit s så som NDVI-värde och höjd. Detta för attfilterara bort eventuella felaktiga punkter som ej är växlighet.Arbetets resultat blev ett punktlager av drygt 21 000 punkter. Punktlagret innehaver informativa attribut i formav höjdvärde, diameter-värde, NDVI-värde, och antalet celler som kronan består av. Noggrannheten vididentifiering av träden är stark beroende av trädens form och placering. Träd med en simpel trädkronaform visarresultat med hög identifieraringsnoggrannhet. Träd med mer komplexa trädkronor (oftast stora lövträd) eller omträden är placerat så att dess trädkronor flyter samman, visar resultaten sämre identifieringsnoggrannhet.Placeringsnoggrannheten beräknades genom en jämförelse med inmätta träd från mätingenjör. Resultatet visaratt placeringen av de identifierade träden har ett medelfel på drygt 2 meter. Höjdvärde uppnår en relativ högnoggrannhet då de är direkt tagna från höjdrastret vars höjdvärde är tagna från lantmäteriets LAS-data med enviss generalisering. Diametervärdena visar låg noggrannhet men med en konstant avvikelse som skulle göravärdena möjliga att använda vid en storleksklassning av trädena. / With help of airborne laser scanning can large amount of LAS-data be collected. Lantmäteriet, a state agencywhose mission is to map Sweden, have done a nationwide collection of LAS-data. This data is available to thepublic and can be downloaded for free from Lantmäteriet. As the collection is nationwide the resolution of theLAS-data is lower than a local collection of LAS-data. The point density of the LAS-data is 1-2 points per squaremeter.This study was done on behalf of Arvika municipality as they wanted to know if it is possible to extract and mapthe urban vegetation in Arvika city from Lantmäteriets LAS-data. What was wanted from Arvika municipality wasa point layer that shows approximately positions of trees over 3-5 meters. The point layer would have attributeswith information about the size of the tree. This would be useful for the municipality’s GIS apartment tosimulate shading when planning projects, modelling trees and producing informative vegetation statistics. Itwould also be useful for the park department which previously lacked data on their urban vegetation.This project was implemented using and combining different functions in ArcGIS Pro. Firstly the LAS-dataset wasconverted to raster format. This raster is then multiplied by -1 which then is inverted. This inverted raster isbeing used in hydrological analysis functions that compares cells with their neighbours. By doing this, a pointlayer showing local height peaks is produced. This point layer is later fed with a height value from the raster anda NDVI-value. This NDVI-value is obtained by using two bands (band 1 and 4) obtained from Lantmäteriet’sorthophoto.A diameter value is calculated by determining the number of cells that make up the crown for the tree. Thisvalue is very generalized but has a relatively constant difference and could therefore be used for sizeclassification of trees. The local height points are also filtered with a couple of criteria that they must meet. Thecriteria are based on some of the informative data that has been produced like NDVI and height value. This is tofilter out any potential incorrect points that are not any form of vegetation.The result of this study consists of just over 21 000 tree points. The point layer has informative attribute whichshows height, diameter, NDVI and the amounts of cells the crown of the tree consists of. The accuracy inidentifying the trees is strongly dependent on the shape and location of the trees. Trees with a simple crownshows results with high identification accuracy. Trees with more complex (most often big deciduous trees) orwhen the trees are located close to each other so that their crowns flow together, it shows results with loweridentification accuracy. The positional accuracy was calculated by a comparison with trees measured by a surveyengineer. The results shows that the positional accuracy of identified trees have a mean deviation of just over 2meters. The height value attains a relatively high accuracy as it is directly taken from the height value inLantmäteriets LAS-dataset, with a certain generalization. The diameter value shows low accuracy but with aconstant difference which could be possible to use for tree size classification of the trees.

No description available.

Digital twins in manufacturing, logistics, retail, and healthcare can help companies makebusiness decisions by simulating changes prior to implementing such changes in real life.In robotic teleoperation, virtual reality technology such as head mounted displays canincrease operator performance. In the mining equipment industry, teleoperation is quitean established concept, using a video feed for visualization, and often similiar or the samecontrol panels as on the real machine. However, cameras don’t provide depth perceptionfor the operator, and the lighting conditions in a mine may make photogrammetry a lessthan ideal solution. Epiroc is currently working on a digital twin simulation softwarein Unity, which could be extended for teleoperation purposes. As a complement to thissoftware, a fast, high-definition Ouster OS0-128 LiDAR was used to render a point cloudof a physical environment. A Unity GameObject script was written in C# that receivesand renders coordinates as a point cloud. Two Python scripts were written to convert theLiDAR data using the Ouster SDK to coordinates, and then sending these coordinates overa TCP connection, either on the same machine, or over Wi-Fi. The Python scripts used twodifferent data formats, and the performance difference between these two data formatswas compared. The results indicated that Wi-Fi transfer of LiDAR data could be a viablesolution to continously scanning the surrounding area of equipment being teleoperatedwith quite a low delay and latency

LiDAR is a remote sensing technology that uses a laser to map the surrounding environment. With its many applications, for example in autonomous vehicles, LiDAR is a growing field within technology and research. The silicon photomultiplier (SiPM) is a solid state device commonly used in the receiving system of a LiDAR. However, ambient light from sun or other light sources is also seen by the photodetector which creates noise in the receiving system. The purpose of this work is to examine how the performance of a 1D LiDAR with a SiPM receiver can be predicted, given a certain level of ambient light, target reflectance and measuring distance. This was carried through by working with mathematical models and comparing the outcomes to lab measurements of a certain LiDAR model. The outcome showed that describing the performance of the particular LiDAR by a model based on incident photon rate was difficult mainly due to the unknown relation between how the voltage signal threshold in the receiving electric circuit for when the LiDAR stops the time measurement relates to the number of microcells activated in the SiPM during a time span. However, the results obtained suggest a threshold value of between around 20 and 60 microcells activated within a 1 ns time interval, but further tests are needed in order to confirm or reject this. Of the two other approaches tried, the signal voltage model gave reasonable results for the values tested but in a rather indirect way. The other approach, describing the connection between DC noise and AC RMS noise in the receiving system gave results deviating between 40% and 320% from the lab results, i.e. not so good match. / LiDAR är en fjärranalysteknik som använder laser för att kartlägga ett geografiskt område. Med flertalet användningsområden, bland annat inom industrin för självstyrande fordon, är LiDAR ett växande teknik- och forskningsområde. Kisel-fotomultiplikatorn är en halvledarapparat som ofta används i LiDAR-mottagarsystemet. Bakgrundsljus från omgivningen detekteras dock också av fotodetektorn vilket orsakar  brus i mottagarsystemet. Detta arbete syftar till att testa metoder för hur prestandan hos en 1D LiDAR med en kisel-fotomultiplikator i mottagaren kan estimeras, i en viss nivå av bakgrundsljus, med en viss målreflektans på ett visst mätavstånd. Detta utfördes genom att arbeta med matematiska modeller och jämföra dess resultat med resultat från laborativa tester på en viss LiDAR-modell. Det visade sig vara svårt att beskriva prestandan för denna LiDAR enligt en modell baserad på fotonflöde, huvudsakligen på grund av den okända kopplingen mellan tröskelnivån i termer av voltsignal i mottagarkretsen då tidmätningen stoppas och antalet aktiverade mikroceller i kisel-fotomultiplikatorn under ett visst tidsintervall. De resultat som erhölls visar dock på en träskelnivå någonstans mellan 20 och 60 mikroceller inom ett 1 ns tidsintervall, men ytterligare tester bör genomföras för att konfirmera eller förkasta detta. Av de övriga två angreppssätten prövade, gav modellen baserad på voltsignal rimliga resultat för värdena testade men på ett relativt indirekt sätt. Försök till beskrivning av sambandet mellan DC brus och AC RMS brus i mottagarsystemet gav resultat med mellan 40% och 320% avvikelse från de laborativa mätningarna, relativt dåligt alltså.

The mining industry uses vehicles for a wide range of applications, including excavation and transportation of rock and soil. Currently, this requires a lot of human labour, mainly drivers, but efforts are being made to increase automation, e.g. using autonomous vehicles. In order for a vehicle to reach any level of autonomy, it needs to be aware of its surroundings, for instance by using sensors. The placement of the sensors is a difficult problem. The goal of this project was to create a tool that would simplify the sensor placement process. The tool should simulate sensors on autonomous vehicles, both by visualizing their field of view and by generating synthetic data. The tool was created, including simulation environments, models of different types of sensors and tools to analyze the results of the simulation. Both the field of view visualization and the data analysis tools were shown to be powerful tools for evaluating sensor placements. All of the sensor models are able to generate data, with different levels of realism. The radar model and the camera model give a good estimation of what the sensors can detect, while the lidar model is capable of generating realistic data. / Gruvindustrin använder fordon till ett stort antal olika uppgifter, bland annat till att gräva ut och förflytta sten och jord. Detta kräver för tillfället mycket manuellt arbete, framförallt förare, men försök att automatisera delar av arbetet utförs, till exempel genom att använda autonoma fordon. För att ett fordon ska kunna bli autonomt krävs det att det kan känna av sin omgivning, exempelvis genom att använda sensorer. Sensorplacering är ett svårt problem. Målet med projektet var att skapa ett verktyg för att förenkla sensorplaceringsprocessen. Verktyget skulle simulera sensorer på autonoma fordon, både genom att visualisera deras synfält och genom att generera syntetisk data. Verktyget skapades, inklusive simuleringsmiljöer, modeller av olika typer av sensorer, och verktyg för att analysera genererad data. Både synfältsvisualiseringen och datagenereringen visades vara användbara verktyg för att utvärdera sensorplaceringar. Alla sensormodellerna kan generera data, med olika realistiska resultat. Radarmodellen och kameramodellen ger en bra uppskattning av vad sensorerna kan detektera, medan lidarmodellen kan generera realistisk data.

In this study, the weather impact on lidar signals has been researched. A lidar system was placed with a target at approximately 90 m and has together with a weather station collected data for about a year before this study. By using the raw detector data from the lidar, the full waveform can be obtained and the amplitude of the return pulse can be calculated. Atmospheric attenuation of lidar signals is often modeled using the lidar equation, which predicts an exponential decrease in energy over the distance. The factor in the exponent is referred to as the extinction coefficient and it is the main property studied in this thesis. By utilizing models for the extinction coefficient under different weather conditions, it is possible to simulate the performance of the lidar.  The extinction coefficient was calculated using different empirical models. The empirical models investigated in this thesis are the Kim and Kruse models for known visibility, the Al Naboulsi model for different types of fog with known visibility, the Carbonneau model for known precipitation amount in rainy conditions, and a similar model for snowy conditions. For the case of rain, a physical model was also used, which is derived through Mie theory. The physical model requires a particle size distribution, which is the number of particles of a certain radius per unit volume. A particle size distribution for rain was generated using the Ulbrich raindrop size distribution, using the precipitation amount recorded by the weather station. Particle size distributions for radiation and advection fog were also simulated.  The measured attenuation in lidar signals was compared to the predicted attenuation that was calculated using different models for the extinction coefficient in the lidar equation. Generally, the models tend to underestimate the amplitude of the return pulse. This can partially be explained by the assumptions used to derive the lidar equation, which neglects all augmentation of the beam. The visibility models gave more accurate results compared to the precipitation models. This was expected, since visibility is defined as a measure of attenuation and precipitation amount is not.  When a lidar signal is emitted, the light will be reflected from optical surfaces within the lidar and cause a pulse to be detected. This pulse is referred to as the zeropulse. In the first couple of meters of the transmission, we expect to see some backscattered light from adverse weather, since the detector has a larger solid angle at shorter distances. This returned light will be combined with the zeropulse and cause it to expand in width. By examining the zeropulse, it was possible to observe a difference between the average zeropulse under some different weather conditions. This leads to the conclusion that it may be possible to extract some information about current weather conditions from the zeropulse data, given that there is little ambient light and snowy weather conditions.  By integrating the zeropulse, variations in the shape of the zeropulse could be described by a single value. Then by separating the data into low and high visibility populations, the zeropulse integral could be used to predict the visibility. The conclusion was that the zeropulse integral can accurately predict whether visibility is above or below a threshold value, given that there is little ambient light and the visibility is known to be below 19950 m.

Light Detection and Ranging (Lidar) is a kind of active sensor that emits a laser pulse and primarily measures the time of flight of the returning pulse and uses it to construct a 3D point cloud of the scene around the lidar sensor. The constructed point cloud is an essential asset for the control of autonomous vehicles, and especially today, an essential basis for the training of autonomous vehicle control models. However, it remains time-consuming, high-risk and expensive to acquire the amounts of data necessary to train the rather complex modern control models. As such, generating the point cloud through simulations becomes a natural solution. Yet, many lidar simulations today produce ideal point clouds, corrected only by random noise, without considering the physical reasons behind the imperfections visible in real lidar point clouds. The aim of this study was to investigate real-time simulation models for disturbances that may cause imperfections in lidar data. From a base investigation of lidar, disturbances were found, models were investigated and finally a real-time implementation of Atmospheric Effects and attenuation from Beam Divergence was evaluated. It was found that the implemented models could produce physically accurate lidar point placement while keeping the computational time low enough for real-time evaluation. However, to achieve correct separation of target hit rates under Atmospheric Effects, as high as 34% of the points had to be dropped. Additionally, the intensity of the return points could not be properly verified. From these results it can be concluded that, with additional verification and adjustment, the presented models can achieve good results for evaluation in real-time. The results of this study thus serve as a support for future developments of realistic real-time lidar simulations, for use in development of autonomous vehicle control models and implementation of digital twins.

The primary focus of this research is evaluation of feasibility, applicability, and accuracy of Doppler Light Detection And Ranging (LIDAR) sensors as non-contact means for measuring track speed, distance traveled, and curvature. Speed histories, currently measured with a rotary, wheel-mounted encoder, serve a number of useful purposes, one significant use involving derailment investigations. Distance calculation provides a spatial reference system for operators to locate track sections of interest. Railroad curves, using an IMU to measure curvature, are monitored to maintain track infrastructure within regulations. Speed measured with high accuracy leads to high-fidelity distance and curvature data through utilization of processor clock rate and left-and right-rail speed differentials during curve navigation, respectively. Wheel-mounted encoders, or tachometers, provide a relatively low-resolution speed profile, exhibit increased noise with increasing speed, and are subject to the inertial behavior of the rail car which affects output data. The IMU used to measure curvature is dependent on acceleration and yaw rate sensitivity and experiences difficulty in low-speed conditions. Preliminary system tests onboard a 'Hy-Rail' utility vehicle capable of traveling on rail show speed capture is possible using the rails as the reference moving target and furthermore, obtaining speed profiles from both rails allows for the calculation of speed differentials in curves to estimate degrees curvature. Ground truth distance calibration and curve measurement were also carried out. Distance calibration involved placement of spatial landmarks detected by a sensor to synchronize distance measurements as a pre-processing procedure. Curvature ground truth measurements provided a reference system to confirm measurement results and observe alignment variation throughout a curve. Primary testing occurred onboard a track geometry rail car, measuring rail speed over substantial mileage in various weather conditions, providing high-accuracy data to further calculate distance and curvature along the test routes. Tests results indicate the LIDAR system measures speed at higher accuracy than the encoder, absent of noise influenced by increasing speed. Distance calculation is also high in accuracy, results showing high correlation with encoder and ground truth data. Finally, curvature calculation using speed data is shown to have good correlation with IMU measurements and a resolution capable of revealing localized track alignments. Further investigations involve a curve measurement algorithm and speed calibration method independent from external reference systems, namely encoder and ground truth data. The speed calibration results show a high correlation with speed data from the track geometry vehicle. It is recommended that the study be extended to provide assessment of the LIDAR's sensitivity to car body motion in order to better isolate the embedded behavior in the speed and curvature profiles. Furthermore, in the interest of progressing the system toward a commercially viable unit, methods for self-calibration and pre-processing to allow for fully independent operation is highly encouraged. / Ph. D.

Recent developments in surveying and mapping technologies have greatly enhanced our ability to model and analyze both outdoor and indoor environments. This research advances the traditional concept of digital twins—static representations of physical spaces—by integrating real-time data on human occupancy and movement to develop a dynamic digital twin. Utilizing the newly constructed mixed-use building at Virginia Tech as a case study, this research leverages 11 terrestrial lidar sensors to develop a dynamic digital model that continuously captures human activities within public spaces of the building. Three distinct object detection methodologies were evaluated: deep learning models, OpenCV-based techniques, and Blickfeld's lidar perception software, Percept. The deep learning and OpenCV techniques analyzed projected 2D raster images, while Percept utilized real-time 3D point clouds to detect and track human movement. The deep learning approach, specifically the YOLOv5 model, demonstrated high accuracy with an F1 score of 0.879. In contrast, OpenCV methods, while less computationally demanding, showed lower accuracy and higher rates of false detections. Percept, operating on real-time 3D lidar streams, performed well but was susceptible to errors due to temporal misalignment. This study underscores the potential and challenges of employing advanced lidar-based technologies to create more comprehensive and dynamic models of indoor spaces. These models significantly enhance our understanding of how buildings serve their users, offering insights that could improve building design and functionality. / Master of Science / Americans spend an average 87% of their time indoors, but mapping these spaces has been a challenge. Traditional methods like satellite imaging and drones do not work well indoors, and camera-based models can be invasive and limiting. By contrast, lidar technology can create detailed maps of indoor spaces while also protecting people's privacy—something especially important in buildings like schools. Currently, most technology creates static digital maps of places, called digital twins, but these do not show how people actually use these spaces. My study aims to take this a step further by developing a dynamic digital twin. This enhanced model shows the physical space and incorporates real-time information about where and how people move within it. For my research, I used lidar data collected from 11 sensors in a mixed-use building at Virginia Tech to create detailed images that track movement. I applied advanced computer techniques, including machine learning and computer vision, to detect human movement within the study space. Specifically, I used methods such as YOLOv5 for deep learning and OpenCV for movement detection to find and track people's movements inside the building. I also compared my techniques with a known software called Percept by Blickfeld, which detects moving objects in real-time from lidar data. To evaluate how well my methods worked, I measured them using traditional and innovative statistical metrics against a standard set of manually tagged images. This way, I could see how accurately my system could track indoor dynamics, offering a richer, more dynamic view of how indoor spaces are used.