Jun Jo

A vision-based lane detection system combining appearance segmentation and tracking of salient points

Reliable lane detection is a key component of autonomous vehicles supporting navigation in urban environments. This paper introduces the GOLDIE(Geometric Overture for Lane Detection by Intersections Entirety) system, a vision-based software architecture that uses an on-board single camera to determine the position of road lanes with respect to the vehicle. We propose an efficient vision-based lane-detection system that combines an appearance-based analysis with salient point tracking. The appearance-based analysis consists of segmenting high contrast areas that fit inside a Region-Of-Interest(ROI) on the frame. The salient point tracker selects interesting points based in a reference line, that guides a dynamic ROI. The tracking ROI look for paint lane marks close to the last lane reference found, where road marks are likely to emerge, in order to maintain the usability of the salient point tracker. The tracking is performed with the Lucas-Kanade algorithm and the lane points candidates are selected according to a predefined triangular model. Once such lanes points are detected, the vehicle position is estimated based on the intersection of linearised lanes determined through a vanishing point approach. Experiments and comparisons with other algorithms illustrate the applicability of the method.

A Usability Study for Signal Strength Based Localisation

The Global Positioning System (GPS) may be one of the largest technological advances of the human beings. This technology became an essential component of most vehicles including airplanes, cars and ships. This device can be used anywhere around the world. While GPS was made for outdoors, there are major problems about indoor localisation. There are some useful indoor localisation tools that have been researched, such as WiFi and ZigBee technology. But due to the inaccuracy as well as interference that can occur in indoor locations, such as walls, other Wireless Sensor Networks (WSNs) and human interference, the recent systems that have been created, are still deemed to be unstable. Thus the increase of precision is required for these indoor systems. This research investigated the relationships between signal strengths and distance in the real situation. The data set obtained was analysed by using a regression method. Sensitivity was studied with Confidence Interval. The outcome of this research is expected to provide the users with a practical guideline for signal strength based localisation.

A Drone-Based Building Inspection System Using Software-Agents

Regular building inspections are a key means of identifying defects before getting worse or causing a building failure. As a tool for building condition inspections, Unmanned Aerial Vehicles (UAVs) or drones offer considerable potential allowing especially high-rise buildings to be visually assessed with economic and risk-related benefits. One of the critical problems encountered in automating the system is that the whole process involves a very complicated and significant amount of computational tasks, such as UAV control, localisation, image acquisition and abnormality analysis using machine learning techniques. Distributed software agents interact and collaborate each other in complicated systems and improve the reliability, availability and scalability. This research introduces a ubiquitous concept of software-agents to a drone-based building inspection system that is applied to crack-detection on concrete surfaces. The architecture and new features of the proposed system will be discussed.

An Efficient Ego-Lane Detection Model to Avoid False-Positives Detection of Guardrails

Detecting lane markings is a challenging task for vision-based systems due to uncontrolled lighting environments present on the roads. Road infrastructures surrounding the painted markings such as guardrails and curbs often reduce the accuracy of existing solutions. The mentioned infrastructure frequently behaves like lane markings increasing the occurrence of candidate features to be selected by lane detectors. Most of the lane detectors use machine learning techniques with long training phases and inflexible models to achieve some level of robustness, therefore an efficient approach capable of performing unsupervised learning is required. The adoption of an efficient model, which can monitor the ego-lane boundaries while identifying false positive references, is discussed in this paper. The proposed architecture allows the combination of multiple image-processing cues to improve accuracy and robustness on vision-based methods. Our method performed with high accuracy in lane marking detection under highly dynamic lighting, including in presence of guardrails.

Multiscale Wavelet Method for Heart Abnormality Detection Within IoTs Environment

The Internet of Things (IoT) is one of the fastest emerging technologies with many different applications in a number of fields. Within the medical domain it is rapidly expanding the capabilities of the IoT technology. The adoption of the IoT to ECG monitoring has the potential to provide maximum information about the electrical activity of the heart as well as allows the large volume of information to be fully used. This paper proposes the idea of a system that utilizes the IoT with the pre-processing and feature extractions done with the use of discrete wavelet transforms and multiscale analysis. However, efficiency is an important issue due to large and complicated interconnections. The use of features rather than raw data makes the process efficient. We introduce multiscale concept based on modulus maxima and minima for feature extraction, which relies on relative distances from R peaks. We named it Gated multiscale selection and also extended this methodology and introduced a Linear multiscale approach. We have found that the specific Linear multiscale combinations achieve the highest accuracy in individual peak identifications and we demonstrated that the proposed method performs better than methods found in literature.

A Comparative Study of Wi-Fi and Bluetooth for Signal Strength-Based Localisation

With the growing need for indoor localisation solutions, this paper investigates the practical applications of wireless networking technologies based on the empirical study. By comparing between the two most widely used wireless technologies, aims to identify which technology, between Wi-fi and Bluetooth, is more capable in RSS-based localisation. Field experiments were conducted in order to collect the data to model the propagation of the two technologies. This study demonstrates that, through comparing the derived models to empirical data, Bluetooth has the potential to improve indoor localisation methods due to its more accurate model.

Detection and Classification of Vehicle Types from Moving Backgrounds

Using unmanned aerial vehicles (UAV) as devices for traffic data collection exhibits many advantages in collecting traffic information. This paper introduces a new vehicle dataset based on image data collected by UAV first. Then a novel learning framework for robust on-road vehicle recognition is presented. This framework starts with conventional supervised learning to create initial training data set. Then a tracking-based online learning approach is applied on consecutive frames to improve the accuracy of vehicle recogniser. Experimental results show that the proposed algorithm exhibits high accuracy in vehicle recognition at different UAV altitudes with different view scopes, which can be used in future traffic monitoring and control in metropolitan areas.

The Use of Machine Learning for Correlation Analysis of Sentiment and Weather Data

The development of the Internet of Things (IoT) drives us to confront, manage and analyse massive and complicated data generated from various sensors. Also, social media have rapidly become very popular and can be considered as important source of data. Twitter on average generates about 6,000 tweets every second, which total to over 500 million tweets per day. Facebook has over 2 billion monthly active users. The individual posts may be trivial, however, the accumulated big data can provide diverse valuable information, which can be also correlated with IoT and enable human sentiment identification of the environment changes. This work proposes a machine learning model for correlation analysis and prediction of weather conditions and social media posts. In experimental evaluation we demonstrate that the Big Data analysis approach and machine learning techniques can be used to analyse and predict sentiment of different weather conditions.

Reusability Quality Metrics for Agent-Based Robot Systems

Programming for robots is generally problem specific and components are not easily reused. Recently there has been push for robotics programming to integrate software engineering principles into the design and development to improve the reusability, however currently no metrics have been proposed to measure this quality in robotics. This paper proposes the use of reusability metrics from Component-Based Software Engineering (CBSE) and Service-Oriented Architecture (SOA) to measure reusability metrics, and finds that they are applicable to modular, agent-based robotics systems through a case study of an example system.