Eftychios Protopapadakis

Deep Convolutional Neural Networks for efficient vision based tunnel inspection

The inspection, assessment, maintenance and safe operation of the existing civil infrastructure consists one of the major challenges facing engineers today. Such work requires either manual approaches, which are slow and yield subjective results, or automated approaches, which depend upon complex handcrafted features. Yet, for the latter case, it is rarely known in advance which features are important for the problem at hand. In this paper, we propose a fully automated tunnel assessment approach; using the raw input from a single monocular camera we hierarchically construct complex features, exploiting the advantages of deep learning architectures. Obtained features are used to train an appropriate defect detector. In particular, we exploit a Convolutional Neural Network to construct high-level features and as a detector we choose to use a Multi-Layer Perceptron due to its global function approximation properties. Such an approach achieves very fast predictions due to the feedforward nature of Convolutional Neural Networks and Multi-Layer Perceptrons.

4D reconstruction of the past

One of the main characteristics of the Internet era we are living in, is the free and online availability of a huge amount of data. This data is of varied reliability and accuracy and exists in various forms and formats. Often, it is cross-referenced and linked to other data, forming a nexus of text, images, animation and audio enabled by hypertext and, recently, by the Web3.0 standard. Search engines can search text for keywords using algorithms of varied intelligence and with limited success. Searching images is a much more complex and computationally intensive task but some initial steps have already been made in this direction, mainly in face recognition. This paper aims to describe our proposed pipeline for integrating data available on Internet repositories and social media, such as photographs, animation and text to produce 3D models of archaeological monuments as well as enriching multimedia of cultural / archaeological interest with metadata and harvesting the end products to EUROPEANA. Our main goal is to enable historians, architects, archaeologists, urban planners and affiliated professionals to reconstruct views of historical monuments from thousands of images floating around the web.

Stacked Autoencoders for Outlier Detection in Over-the-Horizon Radar Signals

Detection of outliers in radar signals is a considerable challenge in maritime surveillance applications. High-Frequency Surface-Wave (HFSW) radars have attracted significant interest as potential tools for long-range target identification and outlier detection at over-the-horizon (OTH) distances. However, a number of disadvantages, such as their low spatial resolution and presence of clutter, have a negative impact on their accuracy. In this paper, we explore the applicability of deep learning techniques for detecting deviations from the norm in behavioral patterns of vessels (outliers) as they are tracked from an OTH radar. The proposed methodology exploits the nonlinear mapping capabilities of deep stacked autoencoders in combination with density-based clustering. A comparative experimental evaluation of the approach shows promising results in terms of the proposed methodologys performance.

Image Based Approaches for Tunnels’ Defects Recognition via Robotic Inspectors

In this paper we present a visual based approach, utilized for the detection of concrete defects in tunnels. The detection mechanism is a hybrid approach, based on both image processing and deep learning models. Initial detections are validated by an expert, in order to create a robust data set in short time, saving resources during annotation process. Then, a deep-learning classifier is trained and applied for the inspection. The fully automated system, performs well, in various environments, and can be, easily, implemented with most robotic systems.

Monocular camera fall detection system exploiting 3d measures: a semi-supervised learning approach

Falls have been reported as the leading cause of injury-related visits to emergency departments and the primary etiology of accidental deaths in elderly. The system presented in this article addresses the fall detection problem through visual cues. The proposed methodology utilize a fast, real-time background subtraction algorithm based on motion information in the scene and capable to operate properly in dynamically changing visual conditions, in order to detect the foreground object and, at the same time, it exploits 3D spaces measures, through automatic camera calibration, to increase the robustness of fall detection algorithm which is based on semi-supervised learning. The above system uses a single monocular camera and is characterized by minimal computational cost and memory requirements that make it suitable for real-time large scale implementations.

Crack Identification Via User Feedback, Convolutional Neural Networks and Laser Scanners for Tunnel Infrastructures

In this paper, a deep learning approach synergetically to a laser scanning process are employed for the visual detection and accurate description of concrete defects in tunnels. Analysis is performed over raw RGB images; Convolutional Neural Network serves as the crack detector, during the inspection. In case of a positive detection, the tunnel’s cross-section morphology is assessed via 3D point clouds, created by a laser scanner, allowing the identification of deformations in the compartment. The proposed approach, in contrast to the existing ones, emphasizes on applicability (easy initialization, no preprocessing of the input data) and provides a holistic assessment of the structure; reconstructed 3D model allows the fast identification of structural divergence from the original design, alerting the engineers for possible dangers.

Semi-supervised vision-based maritime surveillance system using fused visual attention maps

This paper presents a vision-based system for maritime surveillance, using moving PTZ cameras. The proposed methodology fuses a visual attention method that exploits low-level image features appropriately selected for maritime environment, with appropriate tracker, without making any assumptions about environmental or visual conditions. The offline initialization is based on large graph semi-supervised technique. System’s performance was evaluated with videos from cameras placed at Limassol port and Venetian port of Chania. Results suggest high detection ability, despite dynamically changing visual conditions and different kinds of vessels, all in real time.

Semi-Supervised Image Meta-Filtering Using Relevance Feedback in Cultural Heritage Applications

An image filtering scheme for images of cultural interest is presented. The proposed methodology utilize a semi supervised approach for the creation of an appropriate distance learning metric, which is used for the filtering. User’s feedback is involved only for a minor set of data, defined using OPTICS algorithm and sparse modeling representative selection. Such an approach facilitates the refinement of retrieval results, always under the scope of the user needs. The described methodology can be easily implemented for a variety of feature vectors and data sets.

3D measures exploitation for a monocular semi-supervised fall detection system

Falls have been reported as the leading cause of injury-related visits to emergency departments and the primary etiology of accidental deaths in elderly. Thus, the development of robust home surveillance systems is of great importance. In this article, such a system is presented, which tries to address the fall detection problem through visual cues. The proposed methodology utilizes a fast, real-time background subtraction algorithm, based on motion information in the scene and pixels intensity, capable to operate properly in dynamically changing visual conditions, in order to detect the foreground object. At the same time, it exploits 3D space’s measures, through automatic camera calibration, to increase the robustness of fall detection algorithm which is based on semi-supervised learning approach. The above system uses a single monocular camera and is characterized by minimal computational cost and memory requirements that make it suitable for real-time large scale implementations.

4D reconstruction of the past: the image retrieval and 3D model construction pipeline

One of the main characteristics of the Internet era we are living in, is the free and online availability of a huge amount of data. This data is of varied reliability and accuracy and exists in various forms and formats. Often, it is cross-referenced and linked to other data, forming a nexus of text, images, animation and audio enabled by hypertext and, recently, by the Web3.0 standard. Our main goal is to enable historians, architects, archaeolo- gists, urban planners and affiliated professionals to reconstruct views of historical monuments from thousands of images floating around the web. This paper aims to provide an update of our progress in designing and imple- menting a pipeline for searching, filtering and retrieving photographs from Open Access Image Repositories and social media sites and using these images to build accurate 3D models of archaeological monuments as well as enriching multimedia of cultural / archaeological interest with metadata and harvesting the end products to EU- ROPEANA. We provide details of how our implemented software searches and retrieves images of archaeological sites from Flickr and Picasa repositories as well as strategies on how to filter the results, on two levels; a) based on their built-in metadata including geo-location information and b) based on image processing and clustering techniques. We also describe our implementation of a Structure from Motion pipeline designed for producing 3D models using the large collection of 2D input images (>1000) retrieved from Internet Repositories.