A5070626607- Robotics and Sensor-Based Localization
- Advanced Vision and Imaging
- Advanced Image and Video Retrieval Techniques
- Insect Pheromone Research and Control
- Advanced Chemical Sensor Technologies
- Robotics and Automated Systems
- Robotic Path Planning Algorithms
- Optical measurement and interference techniques
- Social Robot Interaction and HRI
- 3D Surveying and Cultural Heritage
- Modular Robots and Swarm Intelligence
- Context-Aware Activity Recognition Systems
- Advanced Neural Network Applications
- Indoor and Outdoor Localization Technologies
- Image and Object Detection Techniques
- Olfactory and Sensory Function Studies
- Remote Sensing and LiDAR Applications
- Advanced Optical Sensing Technologies
- Gas Sensing Nanomaterials and Sensors
- Technology Use by Older Adults
- Image Retrieval and Classification Techniques
- Neurobiology and Insect Physiology Research
- Robot Manipulation and Learning
- Video Surveillance and Tracking Methods
- Air Quality Monitoring and Forecasting
Universidad de Málaga
2016-2025
Centre for Automation and Robotics
2024
Instituto de Investigación Biomédica de Málaga
2016-2022
University of Groningen
2016-2017
Traditional approaches to stereo visual simultaneous localization and mapping (SLAM) rely on point features estimate the camera trajectory build a map of environment. In low-textured environments, though, it is often difficult find sufficient number reliable and, as consequence, performance such algorithms degrades. This paper proposes PL-SLAM, SLAM system that combines both points line segments work robustly in wider variety scenarios, particularly those where are scarce or not...
This paper introduces a dataset gathered entirely in urban scenarios with car equipped one stereo camera and five laser scanners, among other sensors. One distinctive feature of the present is existence high-resolution images grabbed at high rate (20 fps) during 36.8 km trajectory, which allows benchmarking variety computer vision techniques. We describe sensors employed highlight some applications could be benchmarked using work. Both plain text binary files are provided, as well...
Most approaches to visual odometry estimates the camera motion based on point features, consequently, their performance deteriorates in low-textured scenes where it is difficult find a reliable set of them. This paper extends popular semi-direct approach monocular known as SVO [1] work with line segments, hence obtaining more robust system capable dealing both textured and structured environments. The proposed allows for fast tracking segments since eliminates necessity continuously...
In this paper we propose an efficient solution to jointly estimate the camera motion and a piecewise-rigid scene flow from RGB-D sequence. The key idea is perform two-fold segmentation of scene, dividing it into geometric clusters that are, in turn, classified as static or moving elements. Representing dynamic set rigid drastically accelerates estimation, while segmenting parts allows us separate (odometry) rest motions observed scene. resulting method robustly accurately determines...
This paper presents the first method to compute dense scene flow in real-time for RGB-D cameras. It is based on a variational formulation where brightness constancy and geometric consistency are imposed. Accounting depth data provided by cameras, regularization of field imposed 3D surface (or set surfaces) observed instead image plane, leading more geometrically consistent results. The minimization problem efficiently solved primal-dual algorithm which implemented GPU, achieving previously...
Most approaches to stereo visual odometry reconstruct the motion based on tracking of point features along a sequence images. However, in low-textured scenes it is often difficult encounter large set features, or may happen that they are not well distributed over image, so behavior these algorithms deteriorates. This paper proposes probabilistic approach combination both and line segment works robustly wide variety scenarios. The camera recovered through non-linear minimization projection...
The registration of 3D models by a Euclidean transformation is fundamental task at the core many application in computer vision. This problem non-convex due to presence rotational constraints, making traditional local optimization methods prone getting stuck minima. paper addresses finding globally optimal various problems unified formulation that integrates common geometric modalities (namely point-to-point, point-to-line and point-to-plane). renders independent both number nature...
This work presents a simulation framework developed under the widely used Robot Operating System (ROS) to enable validation of robotics systems and gas sensing algorithms realistic environments. The is rooted in principles computational fluid dynamics filament dispersion theory, modeling wind flow 3D real-world scenarios (i.e., accounting for walls, furniture, etc.). Moreover, it integrates different environmental sensors, such as metal oxide photo ionization detectors, or anemometers. We...
Single image calibration is the problem of predicting camera parameters from one image. This importance when dealing with images collected in uncontrolled conditions by non-calibrated cameras, such as crowd-sourced applications. In this work we propose a method to predict extrinsic (tilt and roll) intrinsic (focal length radial distortion) single We parameterization for distortion that better suited learning than directly parameters. Moreover, additional heterogeneous variables exacerbates...
One of the major disadvantages use Metal Oxide Semiconductor (MOS) technology as a transducer for electronic gas sensing devices (e-noses) is long recovery period needed after each exposure. This severely restricts its usage in applications where concentrations may change rapidly, mobile robotic olfaction, allowing sensor forces robot to move at very low speed, almost incompatible with any practical operation. paper describes design new e-nose which overcomes, great extent, such limitation....
This paper presents the Robot-at-Home dataset (Robot@Home), a collection of raw and processed sensory data from domestic settings aimed at serving as benchmark for semantic mapping algorithms through categorization objects and/or rooms. The contains 87,000+ time-stamped observations gathered by mobile robot endowed with rig four RGB-D cameras 2D laser scanner. Raw have been to produce different outcomes also distributed dataset, including 3D reconstructions geometric maps inspected rooms,...
Metal Oxide Semiconductor (MOX) gas transducers are one of the preferable technologies to build electronic noses because their high sensitivity and low price. In this paper we present an approach overcome a certain extent major disadvantages: slow recovery time (tens seconds), which limits suitability applications where sensor is exposed rapid changes concentration. Our proposal consists exploiting double first-order model MOX-based from steady-state output anticipated in real given...
Finding the relative pose between two calibrated views ranks among most fundamental geometric vision problems. It therefore appears as somewhat a surprise that globally optimal solver minimizes properly defined energy over non-minimal correspondence sets and in original space of transformations has yet to be discovered. This, notably, is contribution present paper. We formulate problem Quadratically Constrained Quadratic Program (QCQP), which can converted into Semidefinite (SDP) using...
The integration of Ambient Assisted Living (AAL) frameworks with Socially Assistive Robots (SARs) has proven useful for monitoring and assisting older adults in their own home. However, the difficulties associated long-term deployments real-world complex environments are still highly under-explored. In this work, we first present MoveCare system, an unobtrusive platform that, through a SAR into AAL framework, aimed to monitor, assist provide social, cognitive, physical stimulation houses...
Robots are often equipped with 2D laser-rangefinders (LRFs) and cameras since they complement well to each other. In order correctly combine measurements from both sensors, it is required know their relative pose, that is, solve extrinsic calibration. this paper we present a new approach such problem which relies on the observations of orthogonal trihedrons profusely found as corners in human-made scenarios. Thus, method does not require any specific pattern, turns calibration process fast...
This work addresses the fundamental problem of pose graph optimization (PGO), which is pervasive in context SLAM, and widely known as SE(d)-synchronization mathematical community. Our contribution twofold. First, we provide a novel, elegant, compact matrix formulation maximum likelihood estimation (MLE) for this problem, drawing interesting connections with connection Laplacian object. Second, even though MLE nonconvex computationally intractable general, exploit recent advances convex...
This paper presents a new method for recognizing places in indoor environments based on the extraction of planar regions from range data provided by hand-held RGB-D sensor. We propose to build plane-based map (PbMap) consisting set 3D patches described simple geometric features (normal vector, centroid, area, etc.). world representation is organized as graph where nodes represent and edges connect planes that are close by. structure permits efficiently select subgraphs representing local...
Place recognition is one of the most challenging problems in computer vision, and has become a key part mobile robotics autonomous driving applications for performing loop closure visual SLAM systems. Moreover, difficulty recognizing revisited location increases with appearance changes caused, instance, by weather or illumination variations, which hinders long-term application such algorithms real environments. In this paper we present convolutional neural network (CNN), trained first time...
One of the main open challenges in visual odometry (VO) is robustness to difficult illumination conditions or high dynamic range (HDR) environments. The difficulties these situations come from both limitations sensors and inability perform a successful tracking interest points because bold assumptions VO, such as brightness constancy. We address this problem deep learning perspective, for which we first fine-tune neural network with purpose obtaining enhanced representations sequences VO....