Search

The nonlocal chiral effective theory is applied to investigate the electromagnetic and strange form factors of nucleon. The bubble and tadpole diagrams are included in the calculation. With the contributions from bubble and tadpole diagrams, the obtained electromagnetic form factors are close to the results without these contributions as long as the low energy constants $c_1$ and $c_2$ are properly chosen, while the magnitudes of strange form factors become larger. Both the electromagnetic and strange form factors are still in good agreement with the experimental data., 11 pages,7 figures

Semantic segmentation models are often affected by illumination changes, and fail to predict correct labels. Although there has been a lot of research on indoor semantic segmentation, it has not been studied in low-light environments. In this paper we propose a new framework, LISU, for Low-light Indoor Scene Understanding. We first decompose the low-light images into reflectance and illumination components, and then jointly learn reflectance restoration and semantic segmentation. To train and evaluate the proposed framework, we propose a new data set, namely LLRGBD, which consists of a large synthetic low-light indoor data set (LLRGBD-synthetic) and a small real data set (LLRGBD-real). The experimental results show that the illumination-invariant features effectively improve the performance of semantic segmentation. Compared with the baseline model, the mIoU of the proposed LISU framework has increased by 11.5%. In addition, pre-training on our synthetic data set increases the mIoU by 7.2%. Our data sets and models are available on our project website.

Deep learning-based semantic segmentation models for building delineation face the challenge of producing precise and regular building outlines. Recently, a building delineation method based on frame field learning was proposed by Girard et al. (2020) to extract regular building footprints as vector polygons directly from aerial RGB images. A fully convolution network (FCN) is trained to learn simultaneously the building mask, contours, and frame field followed by a polygonization method. With the direction information of the building contours stored in the frame field, the polygonization algorithm produces regular outlines accurately detecting edges and corners. This paper investigated the contribution of elevation data from the normalized digital surface model (nDSM) to extract accurate and regular building polygons. The 3D information provided by the nDSM overcomes the aerial images’ limitations and contributes to distinguishing the buildings from the background more accurately. Experiments conducted in Enschede, the Netherlands, demonstrate that the nDSM improves building outlines’ accuracy, resulting in better-aligned building polygons and prevents false positives. The investigated deep learning approach (fusing RGB + nDSM) results in a mean intersection over union (IOU) of 0.70 in the urban area. The baseline method (using RGB only) results in an IOU of 0.58 in the same area. A qualitative analysis of the results shows that the investigated model predicts more precise and regular polygons for large and complex structures.

Point clouds serve as the raw material for various models, such as Building Information Models (BIM). In this work, we investigate the reconstruction steps needed to create models that can be utilized directly for agent-based simulations. The input data for the reconstruction is captured with an indoor mobile mapping system. To show the prominence of this idea, we run social distancing and evacuation simulations on the reconstructed models. The simulations are run with multiple agents using a vision-based pedestrian model and A∗-based path finding algorithm. The limitations of this approach are discussed. The video of the simulation is shared with the audience. Link to the video: https://youtu.be/r2D3IxXt7Ls

Forests play a crucial role in storing carbon and are of paramount importance in maintaining global carbon cycle. Assessment of forest biomass at regional and global level is vital for understanding and monitoring health of both tree species and entire cover. Changes in forest biomass are caused by human activities, natural factors and variations in climate. Forest biomass measurement is necessary for gauging the changes in forest ecosystems. Remote sensing is indispensable for mapping forest biophysical parameters. Microwaves are capable of collecting data even in case of cloud cover as the microwaves are of long wavelength. Microwaves help in retrieving scattering information of target. The goal of this research was to map aboveground biomass (AGB) over Barkot forest range in Dehradun, India. The current work focuses on the retrieval of PolInSAR based scattering information for the estimation of aboveground biomass. Radarsat-2 fully Polarimetric C-band data was used for the estimation of AGB in Barkot forest area. A semi-empirical model, which is Extended Water Cloud Model (EWCM) was utilized for AGB estimation. EWCM considers ground-stem interactions. Due to overestimation of volume scattering, polarization orientation angle shift correction was implemented on the PolInSAR pair. Field biomass data was utilized for accuracy assessment. The results show that coefficient of determination (R2) value of 0.47, Root Mean Square Error (RMSE) of 56.18 (t ha−1) and accuracy of 72% was obtained between modelled biomass against field measured biomass. Hence, it can be inferred from the obtained results that PolInSAR technique, in combination with semi-empirical modelling approach, can be implemented for estimating forest biomass.

Often disasters cause structural damages and produce rubble and debris, depending on their magnitude and type. The initial disaster response activity is evaluation of the damages, i.e. creation of a detailed damage estimation for different object types throughout the affected area. First responders and government stakeholders require the damage information to plan rescue operations and later on to guide the recovery process. Remote sensing, due to its agile data acquisition capability, synoptic coverage and low cost, has long been used as a vital tool to collect information after a disaster and conduct damage assessment. To detect damages from remote sensing imagery (both UAV and satellite images) structural rubble/debris has been employed as a proxy to detect damaged buildings/areas. However, disaster debris often includes vegetation, sediments and relocated personal property in addition to structural rubble, i.e. items that are wind- or waterborne and not necessarily associated with the closest building. Traditionally, land cover classification-based damage detection has been categorizing debris as damaged areas. However, in particular in waterborne disaster such as tsunamis or storm surges, vast areas end up being debris covered, effectively hindering actual building damage to be detected, and leading to an overestimation of damaged area. Therefore, to perform a precise damage assessment, and consequently recovery assessment that relies on a clear damage benchmark, it is crucial to separate actual structural rubble from ephemeral debris. In this study two approaches were investigated for two types of data (i.e., UAV images, and multi-temporal satellite images). To do so, three textural analysis, i.e., Gabor filters, Local Binary Pattern (LBP), and Histogram of the Oriented Gradients (HOG), were implemented on mosaic UAV images, and the relation between debris type and their time of removal was investigated using very high-resolution satellite images. The results showed that the HOG features, among other texture features, have the potential to be used for debris identification. In addition, multi-temporal satellite image analysis showed that debris removal time needs to be investigated using daily images, because the removal time of debris may change based on the type of disaster and its location.

Due to the raw images of multi-lens multispectral (MS) camera has significant misregistration errors, performing image registration for band co-registration is necessary. Image matching is an essential step for image registration, which obtains conjugate features on the overlapped areas, and use them to estimate the coefficients of a transformation model for correcting the geometrical errors. However, due to the none-linear intensity of spectral response, performing feature-based image matching (such as SURF) can only obtain only a few conjugate features on cross-band MS images. Different to SURF that extracts local extremum in a multi-scale space and utilizes a threshold to determine a feature, we proposed a normalized SURF (N-SURF) that extracts features on single scale, calculates the cumulative distribution function (CDF) of features, and obtains consistent features from the CDF. In this study, two datasets acquired from Tetracam MiniMCA-12 and Micasense RedEdge Altum are used for evaluating the matching performance of N-SURF. Results show that N-SURF can extract approximately 2–3 times number of features, match more points, and have more efficient than original SURF. On the other hand, with the successful of MS image matching, we can therefor use the conjugates to compute the coefficients of a geometric transformation model. In this study, three transformation models are used to compare the difference on MS band co-registration, i.e. affine, projective, and extended projective. Results show that extended projective model is better than the others as it can compensate the difference of lens distortion and viewpoint, and has co-registration accuracy of 0.3–0.6 pixels.

National mapping agencies (NMAs) have to acquire nation-wide Digital Terrain Models on a regular basis as part of their obligations to provide up-to-date data. Point clouds from Airborne Laser Scanning (ALS) are an important data source for this task; recently, NMAs also started deriving Dense Image Matching (DIM) point clouds from aerial images. As a result, NMAs have both point cloud data sources available, which they can exploit for their purposes. In this study, we investigate the potential of transfer learning from ALS to DIM data, so the time consuming step of data labelling can be reduced. Due to their specific individual measurement techniques, both point clouds have various distinct properties such as RGB or intensity values, which are often exploited for classification of either ALS or DIM point clouds. However, those features also hinder transfer learning between these two point cloud types, since they do not exist in the other point cloud type. As the mere 3D point is available in both point cloud types, we focus on transfer learning from an ALS to a DIM point cloud using exclusively the point coordinates. We are tackling the issue of different point densities by rasterizing the point cloud into a 2D grid and take important height features as input for classification. We train an encoder-decoder convolutional neural network with labelled ALS data as a baseline and then fine-tune this baseline with an increasing amount of labelled DIM data. We also train the same network exclusively on all available DIM data as reference to compare our results. We show that only 10% of labelled DIM data increase the classification results notably, which is especially relevant for practical applications.

Pedestrian tracking is a significant problem in autonomous driving. The majority of studies carries out tracking in the image domain, which is not sufficient for many realistic applications like path planning, collision avoidance, and autonomous navigation. In this study, we address pedestrian tracking using stereo images and tracking-by-detection. Our framework comes in three primary phases: (1) people are detected in image space by the mask R-CNN detector and their positions in 3D-space are computed using stereo information; (2) corresponding detections are assigned to each other across consecutive frames based on visual characteristics and 3D geometry; and (3) the current positions of pedestrians are corrected using their previous states using an extended Kalman filter. We use our tracking-to-confirm-detection method, in which detections are treated differently depending on their confidence metrics. To obtain a high recall value while keeping a low number of false positives. While existing methods consider all target trajectories have equal accuracy, we estimate a confidence value for each trajectory at every epoch. Thus, depending on their confidence values, the targets can have different contributions to the whole tracking system. The performance of our approach is evaluated using the Kitti benchmark dataset. It shows promising results comparable to those of other state-of-the-art methods.

Interactive motion planing and collaborative positioning will play a key role in future autonomous driving applications. For this purpose, the precise reconstruction and pose estimation of other traffic participants, especially of other vehicles, is a fundamental task and will be tackled in this paper based on street level stereo images obtained from a moving vehicle. We learn a shape prior, consisting of vehicle geometry and appearance features, and we fit a vehicle model to initially detected vehicles. This is achieved by minimising an energy function, jointly incorporating 3D and 2D information to infer the model’s optimal and precise pose parameters. For evaluation we use the object detection and orientation benchmark of the KITTI dataset (Geiger et al., 2012). We can show a significant benefit of each of the individual energy terms of the overall objective function. We achieve good results with up to 94.8% correct and precise pose estimations with an average absolute error smaller than 3° for the orientation and 33 cm for position.

Collaborative Positioning (CP) is a networked positioning technique in which different multi-sensor systems (nodes) enhance the accuracy and precision of these navigation solutions by performing measurements or by sharing information (links) between each other. The wide spectrum of available sensors that are used in these complex scenarios bring the necessity to analyze the sensibility of the system to different configurations in order to find optimal solutions. In this paper, we discuss the implementation and evaluation of a simulation tool that allows us to study these questions. The simulation tool is successfully implemented as a plane based localization problem, in which the sensor measurements are fused in a Collaborative Extended Kalman Filter (C-EKF) algorithm with implicit constraints. Using a real urban scenario with three vehicles equipped with various positioning sensors, the impact of the sensor configuration is investigated and discussed by intensive Monte Carlo simulations. The results show the influence of the laser scanner measurements on the accuracy and precision of the estimation, and the increased performance of the collaborative navigation techniques with respect to the single vehicle method.

This paper addresses the problem of unsupervised domain adaptation on the task of pedestrian detection in crowded scenes. First, we utilize an iterative algorithm to iteratively select and auto-annotate positive pedestrian samples with high confidence as the training samples for the target domain. Meanwhile, we also reuse negative samples from the source domain to compensate for the imbalance between the amount of positive samples and negative samples. Second, based on the deep network we also design an unsupervised regularizer to mitigate influence from data noise. More specifically, we transform the last fully connected layer into two sub-layers — an element-wise multiply layer and a sum layer, and add the unsupervised regularizer to further improve the domain adaptation accuracy. In experiments for pedestrian detection, the proposed method boosts the recall value by nearly \(30\,\%\) while the precision stays almost the same. Furthermore, we perform our method on standard domain adaptation benchmarks on both supervised and unsupervised settings and also achieve state-of-the-art results.

Indoor navigation can be a tedious process in a complex and unknown environment. It gets more critical when the first responders try to intervene in a big building after a disaster has occurred. For such cases, an accurate map of the building is among the best supports possible. Unfortunately, such a map is not always available, or generally outdated and imprecise, leading to error prone decisions. Thanks to advances in the laser scanning, accurate 3D maps can be built in relatively small amount of time using all sort of laser scanners (stationary, mobile, drone), although the information they provide is generally an unstructured point cloud. While most of the existing approaches try to extensively process the point cloud in order to produce an accurate architectural model of the scanned building, similar to a Building Information Model (BIM), we have adopted a space-focused approach. This paper presents our framework that starts from point-clouds of complex indoor environments, performs advanced processes to identify the 3D structures critical to navigation and path planning, and provides fine-grained navigation networks that account for obstacles and spatial accessibility of the navigating agents. The method involves generating a volumetric-wall vector model from the point cloud, identifying the obstacles and extracting the navigable 3D spaces. Our work contributes a new approach for space subdivision without the need of using laser scanner positions or viewpoints. Unlike 2D cell decomposition or a binary space partitioning, this work introduces a space enclosure method to deal with 3D space extraction and non-Manhattan World architecture. The results show more than 90% of spaces are correctly extracted. The approach is tested on several real buildings and relies on the latest advances in indoor navigation.

Over the past decades, a special interest has been given to remote-sensing imagery to automate the detection of damaged buildings. Given the large areas it may cover and the possibility of automation of the damage detection process, when comparing with lengthy and costly ground observations. Currently, most image-based damage detection approaches rely on Convolutional Neural Networks (CNN). These are used to determine if a given image patch shows damage or not in a binary classification approach. However, such approaches are often trained using image samples containing only debris and rubble piles. Since such approaches often aim at detecting partial or totally collapsed buildings from remote-sensing imagery. Hence, such approaches might not be applicable when the aim is to detect façade damages. This is due to the fact that façade damages also include spalling, cracks and other small signs of damage. Only a few studies focus their damage analysis on the façade and a multi-temporal approach is still missing. In this paper, a multi-temporal approach specifically designed for the image classification of façade damages is presented. To this end, three multi-temporal approaches are compared with two mono-temporal approaches. Regarding the multi-temporal approaches the objective is to understand the optimal fusion between the two imagery epochs within a CNN. The results show that the multi-temporal approaches outperform the mono-temporal ones by up to 22% in accuracy.

In recent years, the requirements in the industrial production of elongated objects, e.g., aircraft, have been increased. An essential aspect of the production process is the 3D object detection as well as the qualitative assessment of the captured data. On the one hand high accuracy requirements with a 3D standard deviation of σ3D = 1 mm have to be fulfilled, on the other hand an efficient 3D object capturing is needed. In terms of efficiency, kinematic terrestrial laser scanning (k-TLS) has proven its strength in the recent years. It can be seen as an alternative and is even more powerful than to the well established static terrestrial laser scanning (s-TLS). In order to perform a high accurate 3D object capturing with k-TLS, the 3D object capturing of the initial sensor, the (geo-)referencing of the mobile platform, the synchronisation of all sensors and the system calibration, which means the determination of six extrinsic parameters have to be performed with suitable accuracy. Within this contribution we focus on the system calibration. Therefore an approach based on known reference geometries, here planes, is used (Strübing and Neumann, 2013). As a result, the lever arm and boresight angles are determined. Hereby the number as well as the position and orientation of the reference geometries is of importance. Therefore, an optimal arrangement has to be found. Here a sensitive analysis based on uncertainty propagation is used. A selective search of an optimised arrangement is carried out by a genetic algorithm. Within some examples we demonstrate some theoretical aspects and how an optimisation of the reference geometry arrangement can be achieved.