Top PDF DETECTION AND CLASSIFICATION OF POLE-LIKE OBJECTS FROM MOBILE MAPPING DATA

Laser scanners on a vehicle-based mobile mapping system can capture 3D point-clouds of roads and roadside objects. Since roadside objects have to be maintained periodically, their 3D models are useful for planning maintenance tasks. In our previous work, we proposed a method for detecting cylindrical poles and planar plates in a point-cloud. However, it is often required to further classify pole-like objects into utility poles, streetlights, traffic signals and signs, which are managed by different organizations. In addition, our previous method may fail to extract low pole-like objects, which are often observed in urban residential areas. In this paper, we propose new methods for extracting and classifying pole-like objects. In our method, we robustly extract a wide variety of poles by converting point-clouds into wireframe models and calculating cross-sections between wireframe models and horizontal cutting planes. For classifying pole-like objects, we subdivide a pole-like object into five subsets by extracting poles and planes, and calculate feature values of each subset. Then we apply a supervised machine learning method using feature variables of subsets. In our experiments, our method could achieve excellent results for detection and classification of pole-like objects.

Object-based change detection is a powerful analysis tool for remote sensing data, but few studies consider the potential of temporal semivariogram indices for mapping land-cover changes using object-based approaches. In this study, we explored and evaluated the performance of semivariogram indices calculated from remote sensing imagery, using the Normalized Differential Vegetation Index (NDVI) to detect changes in spatial features related to land cover caused by a disastrous 2015 dam failure in Brazil’s Mariana district. We calculated the NDVI from Landsat 8 images acquired before and after the disaster, then created objects by multiresolution segmentation analysis based on post-disaster images. Experimental semivariograms were computed within the image objects and semivariogram indices were calculated and selected by principal component analysis. We used the selected indices as input data to a support vector machine algorithm for classifying change and no-change classes. The selected semivariogram indices showed their effectiveness as input data for object-based change detection analysis, producing highly accurate maps of areas affected by post-dam-failure flooding in the region. This approach can be used in many other contexts for rapid and accurate assessment of such land-cover changes.

An important reason for the remaining classification errors is the wide diversity of agricultural fields and other objects in the images. In images taken early in the spring, there is not much vegetation in fields. Instead, the variation is related, for example, to different soil types, tilling practices, moisture conditions, ditches, and remaining (dead or green) vegetation from the previous season. Some fields can even be covered with floodwater (see Figure 1 d). Such variations become easily classified as changes. This can be a logical classification result, but considering the desired results of the change detection process, these changes are false alarms. Around the edges of fields, false alarms can also be caused by tree canopies above the field and shadows of nearby trees. Shadow points are included in the training data of fields, but all shadows are not correctly classified. Regarding real changes, objects such as roads, non-cultivated field edges and leafless deciduous vegetation can be difficult to distinguish from non-vegetated fields. Some areas that should be bypassed as belonging to fields are actually very similar to some areas that should be detected as changes (see Figure 1 e).

traversability of terrain regions as the robot attempts to drive over them. Also, the system could es- tablish the correspondence between terrain regions in the local neighborhood of the robot and visual features that result from imaging the terrain regions using a standard stereo rig. Finally, that the system could afford to explore the terrain features in its environment without endangering the overall success of its mission was also an assumption. Starting with the observation that traversability is in the most general sense an affordance, the system implemented an on-line learning method that could accurately predict the traversability properties of a complex terrain using a stereo camera, and both geometric features of the terrain and appearance data. By separating the traversability classifier into two different steps, close range and long range, Manduchi et al. (2005) created a novel system that implemented two different algorithms to two different perspectives on the same problem. Us- ing a long-range 3D obstacle detection and terrain color classification, implemented through a color stereo camera based on stereo range measurement and a color-based classification system to label the detected obstacles according to a set of terrain classes appearance, and a single-axis LIDAR for close-range analysis, to allow the system to discriminate between grass and obstacles such as tree trunks or rocks, the system proved viable and robust for unsupervised autonomous navigation in off-road environments. Dang and Hoffmann (2005) created a model that does not need any a pri- ori information about the shape of the observed objects, but relies on the basic assumption that 3D points standing out of the estimated ground-planes are rigid and therefore obstacles. Santana et al. (2011) model introduced a hybrid approach. Large non planar objects were classified as obstacles while on smaller objects the geometrical relationships between neighbor 3D points were considered.

Our SDAT algorithm takes all the detected moving points as in- put regardless false detections. The strengths are: (i) no need of segmentation for the detected moving points; (ii) no need of moving point detection refinement or pre-classification (pedes- trian or not); (iii) partially scanned objects are also retained, to be robust to moderate occlusions and under detections. However, the downside is that false alarms can be raised if a non-pedestrian object’s spatial distribution is constantly similar to a pedestrian’s. The only constraint is the number of points which lie inside of a pedestrian-sized 2D circle. Apparently, more comprehensive and discriminative features should be incorporated into the SDAT al- gorithm to cope with data with many other types of objects, e.g. cars, buses, so that the method can be used for more complex en- vironments. The Tannenstrasse data from Spinello et at. (2011) will be investigated in the future. One drawback of the method is it is time inefficient since the RANSAC step can take more than 10 minutes for a 2 seconds time interval. So it is not suitable for online tracking even the tracklets are associated progressively in an online fashion.

Abstract. As part of the Greenland Ice Mapping Project (GIMP) we have produced three geospatial data sets for the entire ice sheet and periphery. These are (1) a complete, 15 m resolution image mosaic, (2) ice-covered and ice-free ter- rain classification masks, also posted to 15 m resolution, and (3) a complete, altimeter-registered digital elevation model posted at 30 m. The image mosaic was created from a com- bination of Landsat-7 and RADARSAT-1 imagery acquired between 1999 and 2002. Each pixel in the image is stamped with the acquisition date and geo-registration error to facil- itate change detection. This mosaic was then used to manu- ally produce complete ice-covered and ice-free land classifi- cation masks. Finally, we used satellite altimetry and stereo- photogrammetric digital elevation models (DEMs) to en- hance an existing DEM for Greenland, substantially improv- ing resolution and accuracy over the ice margin and periph- ery.

In this work we presented a highly accurate vehicle localization approach using an automotive LiDAR sensor. The automotive data was simulated by sampling data of a highly dense and accurate Mobile Mapping System, which we also used to generate the reference data. In the future, we will use original data, e.g. gathered by a SICK LMS500 laser scanner. The reference map contains height and intensity images with a fixed resolution of 2 cm. These reference images are 2D representations of the point clouds, whereby we only considered static objects. Dynamic objects, like trees, cars or pedestrians were filtered by a change detection and a tree classification algorithm. In the majority of cases both algorithms worked well. In the change detection algorithm false positive detections appeared if two objects were connected, like a bicycle standing at a pole light. False negatives appeared if two similar or the same objects occurred at the same positions. Here a classification approach could improve the results. The tree classification did not work in cases where trees are connected to

The aim of our research project is to enable a reliable localisation pipeline for MM data obtained in urban areas, and to verify existing data sets according to their localisation accuracy in order to economise the acquisition of ground control. Due to apparent differences in the sensor setup and data, two workflows for Mobile Laser Scanning (MLS) and Mobile Mapping Imaging (MMI) are being developed. The common basis is the utilisation of high-resolution aerial nadir and oblique imagery as an external reference to compensate for vertical as well as for horizontal errors. In a first stage, common features between the ground data and aerial nadir imagery are sought. Based on the imprecise, but approximate exterior orientation of the MM data, more reliable and efficient matching techniques can be employed. For instance, a confined search for correspondences and their verification in the other image can be inferred even from coarse orientation parameters. The next stage will be the integration of oblique images into the pipeline to yield common features on the vertical axis in order to better detect errors in height, and to increase the overall number of tie features considerably. Façades and other vertical objects, such as street lights and traffic signs, are potential objects which can be used for that purpose in the future. In a last step, this tie information allows for either a re-computation of the trajectory or, alternatively, an adjustment of the data as such.

Dense point clouds can be collected efficiently from large areas using mobile laser scanning (MLS) technology. Accurate MLS data can be used for detailed 3D modelling of the road surface and objects around it. The 3D models can be utilised, for example, in street planning and maintenance and noise modelling. Utility poles, traffic signs, and lamp posts can be considered an important part of road infrastructure. Poles and trees stand out from the environment and should be included in realistic 3D models. Detection of narrow vertical objects, such as poles and tree trunks, from MLS data was studied. MLS produces huge amounts of data and, therefore, processing methods should be as automatic as possible and for the methods to be practical, the algorithms should run in an acceptable time. The automatic pole detection method tested in this study is based on first finding point clusters that are good candidates for poles and then separating poles and tree trunks from other clusters using features calculated from the clusters and by applying a mask that acts as a model of a pole. The method achieved detection rates of 77.7% and 69.7% in the field tests while 81.0% and 86.5% of the detected targets were correct. Pole-like targets that were surrounded by other objects, such as tree trunks that were inside branches, were the most difficult to detect. Most of the false detections came from wall structures, which could be corrected in further processing.

In order to recognize all the small objects in the whole city, shape features and contextual features are combined to distinguish one object from another (Golovinskiy, et al., 2009). A hierarchical detection method is introduced in (Yang, et al., 2015), the non-ground points are clustered into a series of supervoxels according to the colour and intensity information. Based on a set of predefined rules in a hierarchical order, those segments are classified into different objects. Yu et al. (2015) propose a semi-automated recognition method to extract the street light poles from the Mobile LiDAR data. The non-ground points are first segmented into small clusters through the Euclidean distance clustering method, and a normalized cut based method is employed to further segment the clusters containing more than one object. Then, the pairwise 3D shape context is generated for both the sample objects and the testing objects. At last, the street light poles are extracted through searching the matched testing objects.

To support the FIFA World Cup in Brazil-2014, a great effort was invested towards improving urban infrastructure in the Brazilian cities. One of the aims was to improve the road network. At that point, it was stated that there was not enough information about the exiting urban infrastructure along the streets. One of the main concerns was to map existing poles and distinguish between their uses. In order to study the problem and develop a possible solution, a pilot study was executed within a University Campus, aiming at mapping and classifying poles using only LiDAR point cloud from mobile mapping survey. This paper describes part of the results, concentrating on the classification of pole tops.

CloneCloud [4] and MAUI [5] solve integer linear programming problems with their decision engine, while ThinkAir [9] calculates the averages of historical execution costs on the cloud server and on the mobile device, and chooses the better one. Namboodiri and Ghose [10] also propose an algorithm to determine whether running an application in the cloud is more power efficient. Wolski et al. [11] implement several bandwidth measurement methodologies and compares their performance in grid computing offloading. None of the works [4] [5] [10] [11] makes the offloading decisions based on rich contexts. Several important studies in this area are summarized below.

of thinking – for instance, between geometry, reason and order, or real estate, space and the urban plan. This can be seen in a heuristic that Zook and Graham term ‘DigiPlace’ described as a mixing of ‘mixing of code, data and physical place’ (Zook and Graham, 2007: 1326). DigiPlace has three main compo- nents which characterise its usefulness: automation, individualisation and dynamism. In this way, it also reflects the major changes more generally found in geo- graphic mobile phone applications: an automated, per- sonalised environment, which is constantly updated, sometimes crowd-sourced and needs few skills to read. Daren’s eyes dart about, staring at market stalls, restaurants and grocers nearby. As they bustle with people, he explains that he is trying to find a shop or building near us that he can look up in Google, and it will show him where we are. There is no GPS, so he is using the location search function to negotiate between the database of DigiPlaces on Google Maps and the surrounding streetscape, with its haphazard shops and ramshackle architectures. Central to Daren’s difficulty is that Gage Street is simply too long for him to easily find the precise location. When he inputs the search term ‘Gage Street’ into Google Maps, the pin keeps landing somewhere up the road and so, in this instance, knowing the name of the street does not help us find our coordinate location. Yet, DigiPlace with its ties to Spatial Big Data also relies on particular axioms, a hierarchy and typology of places that do not necessarily take into account cultural and spatial contexts. As this continues, Daren’s inability to locate his own position becomes a complex triangulation between the places he sees, the phone, and the geographic place database that upholds the mapping interface.

Now a days trend tends to be duplication of cores (cf. Figure 5b) in computers and parallel architectures. The first personal computer with a core-duo arrived in 2005 with AMD1 followed by Intel3. In 2006 Sun4 presented its new octo-core called Niagara2. Intel presents last year a 32 in-order x86 cores [46] and Sun recently announce 80 cores computer. Another emerging CPU concept is many-core: the computer dynamically adapts the number of active cores with respect to the user needs. Many-core is useful because when people do not need the entire power of cores, computer turns off some of them. Until now, 3D objects and virtual environments grew up in parallel to processor power, so researchers were continuously looking for improvements on the collision detection algorithms in order to increase their precision, robustness and efficiency [21, 23]. But now, processors power stays roughly constant while virtual environments are more and more sized, so new scientific

Hans Selye[8] was one of the first to popularize the concept of “stress” back in the 1950s. Since then, psychology as well as medicine and popular culture have accepted stress as a negative fact of life. One of the techniques for relieving stress is known as the “Pilot in command” technique. Pilot training involves coping with emergencies. In face of those critical situations, physiological changes occur, which encourage a narrow focus of attention on the “blood rage” necessary for survival. In a crisis, however, a pilot needs precise hand and foot movements, not gross physical strength, and he or she needs clear thinking, not the tunnel vision of rage. As a consequence, the “natural” survival skills triggered by an emergency can actually lead to a pilot losing control of the aircraft.

End users have appreciated the various possibilities to visualise the results in near-real time, but one common requirement is still to be able to use individually selected printouts of the data. For this reason a new print service is under development enabling a user of a web-client to produce a high-quality and high-resolution (300 dpi) map product in various formats on- demand. The service is based on map templates, consumes several web services and allows for an automatic configuration of a map product, including the positioning and arrangement of map frame elements (e.g., legend, grids, map marginalia). The map configuration is straightforward using the current display extent and information layers selected by a web-client user. A first demonstrator of the print-service was successfully tested during the exercise in Lehnin and frequently requested by the exercise management team and visitor delegations. Experiences from the campaign showed that the first analogue map products were available 25 - 30 minutes after image data acquisition. If provided as OGC-compliant web-services, further information layer can be used for map creation, e.g., information on road traffic and parking occupancy as derived from aerial imagery and terrestrial cameras.

In this section we are discussing the main issues and findings in this project, although the results suggest that gamification actually motivates users to perform indoor mapping more that if they would use the non-gamified version, the main problem is that the results are not coming from objective datasets but subjective ones due to the methods of evaluation, while in related works as Urbanopoly (Celino, et al., 2012) or CityExplorer (Matyas, et al., 2008) were evaluated for around one month collecting data trough logs and by the application data itself. A similar method of evaluation was planned to be applied in this thesis project and then by having two applications available into the wild, gamified and non-gamified one for anyone to download it and later analyze the data uploaded into the database as well as the logs capturing the user behavior. Nevertheless was not possible to reach a fully ideal scenario because of some limitations related with the original application, those limitations and problems are discussed in the next subsections.

Abstract: The paper presents the numerical studies for damage detection in beam structure with mode shape curvatures and its spatial wavelet transform. A small simulated perturbation in the form of transverse slots to be treated as damage in beams and a three stage damage detection process for amplifying the discontinuityis proposed here. Vibration data obtained from the perturbed system is processed for mode shapes which are converted into mode shape curvatures and subsequently fed to the wavelet transform. The study revealsthat the proposed transformation is better in sensitizing damage-induced features than the classical approach based alone on bare modal data. It is observed that the decomposition of the spatial signal into wavelet details can identify the damage position in beam like structure by showing relatively larger peaks at the position of the damage.

Some recent papers have focused on the question of offloading cellular traffic in WLAN networks. In [8] the authors carry out a quantitative study of the performance of 3G mobile data offloading through WiFi networks. They recruited about 100 iPhone users from metropolitan areas and collected statistics on their WiFi connectivity during a period of about two and half weeks. Their findings led them to conclude that offloading this traffic collected is an effective means of accommodating both the current and future growth in traffic. The main difference between this proposed work and [8] is the capacity crunch which is analysed to determine the performance of both WiFi, and Femtocells.

In terms of coarse mode effects, Rome presents similar- ities with Beijing and Kanpur (e.g., Fig. 3). However, the fine mode is never observed to grow as in Beijing, and AOT mostly remains a factor of 2–4 times lower than in these two cities. The Rome AERONET station is located at the outskirts of the city, in a region where Saharan dust advec- tion is recorded ∼20–30% of the time (Barnaba and Gobbi, 2004). Measurements clustering at higher Angstrom coeffi- cients than in Beijing also denote a smaller size of the fine aerosols and less a persistent coarse mode. As expected from the MODIS observations (Barnaba and Gobbi, 2004), in spite of being only 400 km North of Rome, Ispra shows almost no sign of dust impact on its record. Conversely, fine mode growth is more evident at this site in part due to air stagna- tion in a mountain-surrounded valley (the Po valley is one of Europe’s most polluted regions, e.g., Melin and Zibordi, 2005). In fact, high-pollution locations such as Ispra, Mex- ico City and GSFC have their measurements clustering in the α∼1.5, δα ∼−0.5 region, with growing AOT linked to both coagulation-aging and hydration-type increase in R f . It is