Marcin Luckner

Malfunction Immune Wi–Fi Localisation Method

Indoor localisation systems based on a Wi–Fi local area wireless technology bring constantly improving results. However, the whole localisation system may fail when one or more Access Point (AP) malfunctions. In this paper we present how to limit the number of observed APs and how to create a malfunction immune localisation method. The presented solutions are an ensemble of random forests with an additional malfunction detection system. The proposed solution reduces a growth of the localisation error to 4 percent for the floor detection inside a six floor building and 2 metres for the horizontal detection in case of a gross malfunction of an AP infrastructure. The system without proposed improvements may give the errors greater than 30 percent and 7 metres respectively in case of not detected changes in the AP’s infrastructure.

3D model reconstruction and evaluation using a collection of points extracted from the series of photographs

This work describes the whole process of 3D model reconstruction. It begins with the representation of the method that is used to find the matching between photographs and the methodology to use the data to form the initial structure of the reconstructed model, represented by a point cloud. As a next stage, a refinement process is performed, using the bundle adjustment method. A set of stereovision methods is used later on to find a more detailed solution. Those algorithms use pairs of images, therefore as a prerequisite a set of routines that aggregates those results is studied. The paper is concluded with a description of how the point cloud is processed, including the surface reconstruction, to form the result. The described methodology is illustrated with reconstructions of three series of professional photographs from a public repository and one series of amateur photographs created especially for this work. The results were evaluated by the proposed area matching and contour matching measures.

Modified Random Forest Algorithm for Wi–Fi Indoor Localization System

The paper presents a modification of Random Forest approach to the indoor localization problem. The localization solution is based on RSS (Received Signal Strength) from multiple sources of Wi–Fi signal. We analyze two localization models. The first one is built using a straightforward application of a random forest method. The second model is a combination of localization models built for each Access Point from the building’s network using similar technique (Random Forests) as for the first model. The modification proposed in the second model gives us a substantial accuracy improvement when compared to the first model. We test also the solution against a network malfunction when some Access Points are turned off as the malfunction immunity is another important feature of the presented localization solution.

Indoor Localisation Based on GSM Signals: Multistorey Building Study

Among the accurate indoor localisation systems that are using WiFi, Bluetooth, or infrared technologies, the ones that are based on the GSM rely on a stable external infrastructure that can be used even in an emergency. This paper presents an accurate GSM indoor localisation system that achieves a median error of 4.39 metres in horizontal coordinates and up to 64 percent accuracy in floor prediction (for 84 percent of cases the floor prediction is mistaken by not more than a single floor). The test and reference measurements were made inside a six-floor academic building, with an irregular shape, whose dimensions are around 50 metres by 70 metres. The localisation algorithm uses GSM signal readings from the 7 strongest cells available in the GSM standard (or fewer, if fewer than 7 are available). We estimate the location by a three-step method. Firstly, we propose a point localisation solution (i.e., localisation based on only one measurement). Then, by applying the central tendency filters and the Multilayer Perceptron, we build a localisation system that uses a sequence of estimations of current and past locations. We also discuss major accuracy factors such as the number of observed signals or the types of spaces in the building.

Publication of geodetic documentation center resources on internet

Geodetic Documentation Centers collect geodetic and cartographic resources. The resources include spatial data and their metadata. European Union INSPIRE directive imposes an obligation on GDC to publish selected data in the Internet. In this paper, an adequate form of publication is discussed on the base of iGeoMap application. The Internet application iGeoMap merges data from various resources. Depending on the data to present, different types of resources are used. The application can publish spatial data from files (text or binary), a database specialized in spatial data service (PostgreSQL, ORACLE), or web services (Web Map Service, Web Feature Service). Utilization of various data sources by the application is presented in this paper. As a part of the subject, searches of the most popular data (parcels, address points, and control points) are discussed. Various data sources and searching mechanisms involved by the searches in iGeoMap are presented in use cases.

Multiclass SVM classification using graphs calibrated by similarity between classes

In this paper new learning structures, similarity between classes based trees and directed acyclic graph, are presented. The proposed structures are based on a distribution of recognized classes in a data space, unlike the known graph methods such as the tree based One-Against-All (OAA) algorithm or the directed acyclic graph based One-Against-One (OAO) algorithm. The structures are created by grouping similar classes. The similarity between classes is estimated by a distance between classes. The OAO strategy is implemented only for the nearest classes. In other cases the OAA strategy is used. This method allows reduction of the classification costs without a significant growth of the classification error. Algorithms, which create similarity based trees and directed acyclic graph are presented in this paper. These methods are also compared in digits recognition task with existing ones.

Reducing number of classifiers in DAGSVM based on class similarity

Support Vector Machines are excellent binary classifiers. In case of multi-class classification problems individual classifiers can be collected into a directed acyclic graph structure DAGSVM. Such structure implements One-Against-One strategy. In this strategy a split is created for each pair of classes, but, because of hierarchical structure, only a part of them is used in the single classification process. The number of classifiers may be reduced if their classification tasks will be changed from separation of individual classes into separation of groups of classes. The proposed method is based on the similarity of classes. For near classes the structure of DAG stays immutable. For the distant classes more than one is separated with a single classifier. This solution reduces the classification cost. At the same time the recognition accuracy is not reduced in a significant way. Moreover, a number of SV, which influences on the learning time will not grow rapidly.

Application of XGBoost Algorithm in Fingerprinting Localisation Task

An Indoor Positioning System (IPS) issues regression and classification challenges in form of an horizontal localisation and a floor detection. We propose to apply the XGBoost algorithm for both tasks. The algorithm uses vectors of Received Signal Strengths from Wi–Fi access points to map the obtained fingerprints into horizontal coordinates and a current floor number. The original application schema for the algorithm to create IPS was proposed. The algorithm was tested using real data from an academic building. The testing data were split into two datasets. The first data set contains signals from all observed access points. The second dataset consist of signals from the academic network infrastructure. The second dataset was created to eliminate temporary hotspots and to improve a stability of the positioning system. The tested algorithm got similar results as reference methods on the wider set of access points. On the limited set the algorithm obtained the best results.

Long Term Analysis of the Localization Model Based on Wi-Fi Network

The paper presents the analysis of long term accuracy of the localization solution based on Wi-Fi signals. The localization model is built using random forest algorithm and it was tested using data collected between years 2012–2014 inside of a six floor building.

Braille Score

The paper presents a developing computer program that helps the blind people dealing with music notation. The program enables the full path of music processing: starting with a printed musical score and ending with MIDI file which can be performed by an electronic instrument. The recognition module based on an advanced artificial intelligence technology is an engine of the system. Recognized scores are converted into a special internal representation that allows conveying all niceties of music. A record can be also processed with an editor module that is particularly projected for the blind people