Search results for "Computer vision"
showing 10 items of 2353 documents
Multi-Scale Feature Extraction for Vehicle Detection Using Phis-Lbp
2018
International audience; Multi-resolutionobjectdetectionfacesseveraldrawbacksincludingitshighdimensionalityproducedby a richer image representation in different channels or scales. In this paper, we propose a robust and lightweight multi-resolution method for vehicle detection using local binary patterns (LBP) as channel feature. Algorithm acceleration is done using LBP histograms instead of multi-scale feature maps and by extrapolating nearby scales to avoid computing each scale. We produce a feature descriptor capable of reaching a similar precision to other computationally more complex algorithms but reducing its size from 10 to 800 times. Finally, experiments show that our method can obt…
Towards automated and operational forest inventories with T-Lidar
2011
International audience; Forest inventory automation has become a major issue in forestry. The complexity of the segmentation of 3D point cloud is due to mutual occlusion between trees, other vegetation, or branches. That is why, the applications done until now are limited to the estimation of the DBH (Diameter at Breast Height), the tree height and density estimation. Furthermore other parameters could also be detected, such as volume or species of trees (Reulke and Haala) . . . This paper presents an effective approach for automatic detection, isolation of trees and DBH estimation. Tree isolation is achieved using an innovative approach based on a clustering methodology followed by a skele…
A Framework for Mesh Segmentation and Annotation using Ontologies
2015
International audience; Mesh segmentation and annotation using semantics has received an increased interest with the recent democratisation of 3D reconstruction methods. The common approach is to perform this task in two steps, by first segmenting the mesh and then annotating it. However, this approach does not allow one part to take advantage of the other. In image processing, some methods are combining segmentation and annotation, but they are not generic and require implementation adjustments or rewritings for each modification of the expert knowledge. In this work, we describe an original framework that mixes segmen-tation and annotation while minimizing the required geometric analysis …
Depth-of-Field of the Accommodating Eye
2014
Our eyes project information from a three-dimensional world to a basically two-dimensional surface, corresponding to the photoreceptor plane in the retina. In theory, only one plane or surface of world can be in focus at one time. However, the eye exhibits a certain tolerance to out-of-focus images, a feature that is known as depth-of-focus (DOF). The corresponding distance range in which the objects are seen “clearly” is known as depth-of-field (DOFi). Although DOF and DOFi refer to an interval of distances or a dioptric range in the image and object space, respectively, both parameters define a similar concept and are usually interchangeable. This article will mainly refer to DOFi because…
A novel pilot study of automatic identification of EMF radiation effect on brain using computer vision and machine learning
2020
Abstract Electromagnetic field (EMF) radiations from mobile phones and cell tower affect brain of humans and other organisms in many ways. Exposure to EMF could lead to neurological changes causing morphological or chemical changes in the brain and other internal organs. Cellular level analysis to measure and identify the effect of mobile radiations is an expensive and long process as it requires preparing the cell suspension for the analysis. This paper presents a novel pilot study to identify changes in brain morphology under EMF exposure considering drosophila melanogaster as a specimen. The brain is automatically segmented, obtaining microscopic images from which discriminatory geometri…
Orientation Adaptive Minimal Learning Machine for Directions of Atomic Forces
2021
Machine learning (ML) force fields are one of the most common applications of ML in nanoscience. However, commonly these methods are trained on potential energies of atomic systems and force vectors are omitted. Here we present a ML framework, which tackles the greatest difficulty on using forces in ML: accurate prediction of force direction. We use the idea of Minimal Learning Machine to device a method which can adapt to the orientation of an atomic environment to estimate the directions of force vectors. The method was tested with linear alkane molecules. peerReviewed
SHREC'12 Track: 3D mesh segmentation
2012
International audience; 3D mesh segmentation is a fundamental process in many applications such as shape retrieval, compression, deformation, etc. The objective of this track is to evaluate the performance of recent segmentation methods using a ground-truth corpus and an accurate similarity metric. The ground-truth corpus is composed of 28 watertight models, grouped in five classes (animal, furniture, hand, human and bust) and each associated with 4 ground-truth segmentations done by human subjects. 3 research groups have participated to this track, the accuracy of their segmentation algorithms have been evaluated and compared with 4 other state-of-the-art methods.
The central control of oriented locomotion in insects - towards a neurobiological model
2010
A neurobiological model for the higher control of oriented locomotion in insects is developed here. It is based on studies in a genetic model insect, the fruit fly Drosophila melangaster. Reference is made to other insects, where respective functions are not yet studied in Drosophila. Flies can walk swiftly with up to 16 steps per second, each leg, on almost any substrate and in any orientation to the gravity vector. To bring about this astounding maneuverability and agility the central complex within the insect brain derives directional commands for the leg coordination centers in the thoracic nervous system. Neuroanatomical information is taken into account with regard to the general flow…
Convergence Analysis of Distributed Set-Valued Information Systems
2016
This paper focuses on the convergence of information in distributed systems of agents communicating over a network. The information on which the convergence is sought is not rep- resented by real numbers, as often in the literature, rather by sets. The dynamics of the evolution of information across the net- work is accordingly described by set-valued iterative maps. While the study of convergence of set-valued iterative maps is highly complex in general, this paper focuses on Boolean maps, which are comprised of arbitrary combinations of unions, intersections, and complements of sets. For these important class of systems, we provide tools to study both global and local convergence. A distr…
Learning-based multiresolution transforms with application to image compression
2013
In Harten's framework, multiresolution transforms are defined by predicting finer resolution levels of information from coarser ones using an operator, called prediction operator, and defining details (or wavelet coefficients) that are the difference between the exact and predicted values. In this paper we use tools of statistical learning in order to design a more accurate prediction operator in this framework based on a training sample, resulting in multiresolution decompositions with enhanced sparsity. In the case of images, we incorporate edge detection techniques in the design of the prediction operator in order to avoid Gibbs phenomenon. Numerical tests are presented showing that the …