António M. R. Sousa

Blind navigation support system based on Microsoft Kinect

This paper presents a system which extends the use of the traditional white cane by the blind for navigation purposes in indoor environments. Depth data of the scene in front of the user is acquired using the Microsoft Kinect sensor which is then mapped into a pattern representation. Using neural networks, the proposed system uses this information to extract relevant features from the scene, enabling the detection of possible obstacles along the way. The results show that the neural network is able to correctly classify the type of pattern presented as input.

Measuring displacement fields by cross-correlation and a differential technique: experimental validation

A digital image correlation (DIC) algorithm for displacement measurements combining cross-correlation and a differential technique was validated through a set of experimental tests. These tests consisted of in-plane rigid-body translation and rotation tests, a tensile mechanical test, and a mode I fracture test. The fracture mechanical test, in particular, was intended to assess the accuracy of the method when dealing with discontinuous displacement fields, for which subset-based image correla- tion methods usually give unreliable results. The proposed algorithm was systematically compared with the Aramis® DIC-2D commercial code by processing the same set of images. When processing images from rigid-body and tensile tests (associated with continuous displacement fields), the two methods provided equivalent results. When processing images from the fracture mechanical test, however, the proposed method obtained a better qualitative description of the discontinuous displace- ments. Moreover, the proposed method gave a more reliable estimation of both crack length and crack opening displacement of the fractured spe- cimen.©

Vineyard properties extraction combining UAS-based RGB imagery with elevation data

ABSTRACT To differentiate between canopy and vegetation cover is particularly challenging. Nonetheless, it is pivotal in obtaining the exact crops’ vegetation when using remote-sensing data. In this article, a method to automatically estimate and extract vineyards’ canopy is proposed. It combines vegetation indices and digital elevation models – derived from high-resolution images, acquired using unmanned aerial vehicles – to differentiate between vines’ canopy and inter-row vegetation cover. This enables the extraction of relevant information from a specific vineyard plot. The proposed method was applied to data acquired from some vineyards located in Portugal’s north-eastern region, and the resulting parameters were validated. It proved to be an effective method when applied with consumer-grade sensors, carried by unmanned aerial vehicles. Moreover, it also proved to be a fast and efficient way to extract vineyard information, enabling vineyard plots mapping for precision viticulture management tasks.

Multi-Temporal Analysis of Forestry and Coastal Environments Using UASs

Due to strong improvements and developments achieved in the last decade, it is clear that applied research using remote sensing technology such as unmanned aerial vehicles (UAVs) can provide a flexible, efficient, non-destructive, and non-invasive means of acquiring geoscientific data, especially aerial imagery. Simultaneously, there has been an exponential increase in the development of sensors and instruments that can be installed in UAV platforms. By combining the aforementioned factors, unmanned aerial system (UAS) setups composed of UAVs, sensors, and ground control stations, have been increasingly used for remote sensing applications, with growing potential and abilities. This paper’s overall goal is to identify advantages and challenges related to the use of UAVs for aerial imagery acquisition in forestry and coastal environments for preservation/prevention contexts. Moreover, the importance of monitoring these environments over time will be demonstrated. To achieve these goals, two case studies using UASs were conducted. The first focuses on phytosanitary problem detection and monitoring of chestnut tree health (Padrela region, Valpacos, Portugal). The acquired high-resolution imagery allowed for the identification of tree canopy cover decline by means of multi-temporal analysis. The second case study enabled the rigorous and non-evasive registry process of topographic changes that occurred in the sandspit of Cabedelo (Douro estuary, Porto, Portugal) in different time periods. The obtained results allow us to conclude that the UAS constitutes a low-cost, rigorous, and fairly autonomous form of remote sensing technology, capable of covering large geographical areas and acquiring high precision data to aid decision support systems in forestry preservation and coastal monitoring applications. Its swift evolution makes it a potential big player in remote sensing technologies today and in the near future.

Open-Source Indoor Navigation System Adapted to Users with Motor Disabilities☆

Abstract This paper describes the development of a mobile indoor navigation system, supported by a GIS and built using only open source tools. For the sake of simplicity a single building was chosen for the tests converting the floors to digital information from paper plans. The rooms geometry was saved on a proper database with all the adjacent information associated, which can in turn be provided to the clients application by APIs and Web Services. The system is able to calculate the most adequate path between any of the rooms taking into account the user profile which is defined by its degree of mobility (eg. wheelchair). By reading a QR code placed in key places inside the building the user can obtain, on a mobile phone, his current position and receive orientations to any room that he might want to go. The directions hints are complemented with the presentation of real pictures associated to key locations in the path to validate that the correct path is taken by the user.

UAS-based photogrammetry of cultural heritage sites: a case study addressing Chapel of Espírito Santo and photogrammetric software comparison

The cost-effectiveness of unmanned aerial systems (UAS) makes them suitable platforms to survey cultural heritage sites. Developments in photogrammetry provide methods capable to generate accurate 3D models out of 2D aerial images. Considering the involved technologies, the purpose of this paper is to document the Chapel of Espiríto Santo: a very relevant monument for Vila Real (Portugal) that is currently located at the campus of the University of Trás-os-Montes and Alto Douro. The UAS based aerial imagery survey approach is presented along with photogrammetric process to build chapels 3D model. Moreover, two photogrammetric software were compared - Pix4Dmapper Pro and Agisoft Photoscan - in terms of modelling accuracy and functionalities ease of use.

Machine learning classification methods in hyperspectral data processing for agricultural applications

In agricultural applications hyperspectral imaging is used in cases where differences in spectral reflectance of the examined objects are small. However, the large amount of data generated by hyperspectral sensors requires advance processing methods. Machine learning approaches may play an important role in this task. They are known for decades, but they need high volume of data to compute accurate results. Until recently, the availability of hyperspectral data was a big drawback. It was first used in satellites, later in manned aircrafts and data availability from those platforms was limited because of logistics complexity and high price. Nowadays, hyperspectral sensors are available for unmanned aerial vehicles, which enabled to reach a high volume of data, thus overcoming these issues. This way, the aim of this paper is to present the status of the usage of machine learning approaches in the hyperspectral data processing, with a focus on agriculture applications. Nevertheless, there are not many studies available applying machine learning approach to hyperspectral data for agricultural applications. This apparent limitation was in fact the inspiration for making this survey. Preliminary results using UAV-based data are presented, showing the suitability of machine learning techniques in remote sensed data.

UAS-based imagery and photogrammetric processing for tree height and crown diameter extraction

Advances in Unmanned Aerial Systems (UAS) allowed them to become both flexible and cost-effective. When combined with computer vision data processing techniques they are a good way to obtain high-resolution imagery and 3D information. As such, UAS can be advantageous both for agriculture and forestry areas, where the need for data acquisition at specific times and within a specific time frame is crucial, enabling the extraction of several measurements from different crop types. In this study a low-cost UAS was used to survey an area mainly composed by chestnut trees (Castanea sativa Mill.). Flights were performed at different heights (ranging from 30 to 120 m), in single and double grid flight patterns, and photogrammetric processing was then applied. The obtained information consists of orthophoto mosaics and digital elevation models which enable the measurement of individual trees parameters such as tree crown diameter and tree height. Results demonstrate that despite its lower spatial resolution, data from single grid flights carried out at higher heights provided more reliable results than data acquired at lower flight heights. Higher number of images acquired in double grid flights also improved the results. Overall, the obtained results are encouraging, presenting a R2 higher than 0.9 and an overall root mean square error of 44 cm.

Traffic Sign Recognition for Autonomous Driving Robot

This paper introduces a fast Traffic Sign Recognition system developed for a robot, participant in the Autonomous Driving Competition in the Portuguese Festival of Robotics. The Autonomous Driving Robot performs detection and classification of traffic signs and traffic lights based on the analysis of images acquired by a camera mounted on its chassis. The proposed algorithm is composed of three processing stages: detection, pictogram extraction and classification. After the two firsts processing stages, a binary pattern matrix is obtained by color segmentation. In the classification stage two different neural networks were trained to recognize the traffic signs or the traffic light sign. Experimental results show that the system precision is very close to 100% whereas recall presents values above 90% in most of the signs. The proposed system also proves to be reliable and suitable for real-time processing.