Figure 1 - uploaded by Nicola Ferrier
Content may be subject to copyright.
Source publication
Addresses the problem of recovering 3D geometry using an active stereo vision system. Calibration procedures can be adapted to the active stereo configuration, however, considerable effort is required to accurately model and calibrate the kinematics to avoid poor reconstruction. In the active stereo case there will also be errors due to uncertainty...
Similar publications
The objective of this paper is to propose an innovative visual servoing method in order to improve the 3D reconstruction of objects for quantitative measurements. The method uses a Stereovision system that allows us to obtain several shots of an object, at regular intervals according to a predefined trajectory. In our case, the stereo rig is equipp...
Citations
... Thus, the 3D reconstruction from stereo vision gains high importance. And hence there are research theories related to this topic, however, less effort has been made into their practical implementation [8,[13][14][15][16][17][18]. ...
... Biology is one of them. Like robots become smarter by biological inspiration [1] or robot vision systems are built based on a biological model [2], the vision techniques used in robotics can be employed in the research of biology. An example is the automatic visual monitoring of worms, which is discussed in this paper. ...
In this paper, the application issues of vision sensing techniques to analysing the body postures of worms are discussed. The worm targeted here is the nematode C. elegans (short for Caenorhabditis elegans), which is widely used for pursuing research in developmental biology and neurology. The image of C. elegans is firstly segmented based on pixel grey-values, and the extracted worm region is skeletonized. The worm body shape is then modelled with a chain of multiple links. As the link connection is in the same structure of a robot manipulator, a variety of analysis techniques developed in robotics can be readily employed in the automatic analysis of worm images.
... Establish the calibration pattern and collect data, what points on the pattern are to be used as calibration points, is a sub-problem of the camera calibration process [5]. A popular method to collect data is to use a flat printed panel or by observing a calibration object whose geometry in 3D space is known. ...
A neural network-based camera calibration method is proposed for mobile robot equipped with a CMOS vision module. Camera calibration is the process of determining the coordinate relationship between a camera image and its real world space. Accurate camera calibration is required for achieving accurate visual measurements. In order to localize and navigate the robot using vision information, the camera has to be calibrated first. We calibrate the camera using neural networks, which can successfully find the relationship between image coordinates without explicitly calculating all physical camera parameters, including position, orientation, focal length, and lens distortion. Applying the camera calibration as a preprocessing step for the robot localization process is showed an enactment in the robot pose error.
... Data was collected using the method outlined in Neubert et. al. [10] . The model performed well on the data gathered, especially in the x and y directions (seetable 1). ...
... Data was collected using the method outlined in Neubert et. al. [10]. ...
We present a calibration procedure to determine the kinematic
parameters of an active stereo system in a robot-centric frame of
reference. Our goal was to obtain a solution of sufficient accuracy that
the kinematic model information can be used to estimate scene structure
given the measured motion/position of the eye and target object's image
location. We formulate our problem using canonical coordinates of the
rotation group. which enables a particularly simple closed form
solution. Additionally, this formulation and solution provides
quantitative measures of the resulting solutions. Experiments verify
that the solutions are accurate, 3D structure can be estimated from the
kinematic models, and the algorithm can indicate when image errors are
large enough to produce unreliable results
... Data was collected using the method outlined in Neubert et. al. [10]. ...
We present a calibration procedure to determine the kinematic parameters of an active stereo system in a robot-centric frame of reference. Our goal was to obtain a solution of sufficient accuracy that the kinematic model information can be used to estimate scene structure given the measured motion/position of the eye and target object's image location. We formulate our problem using canonical coordinates of the rotation group, which enables a particularly simple closed form solution. Additionally, this formulation and solution provides quantitative measures of the resulting solutions. Experiments verify that the solutions are accurate, 3D structure can be estimated from the kinematic models, and the algorithm can indicate when image errors are large enough to produce un-reliable results.
... Finding these parameters requires a complicated calibration procedure and the collection of several different images [Knight andReid, 2000, Guse, 1999]. Neubert, et al [2001] suggest that a neural network (NN) may provide a more accurate and robust method for 3D reconstruction. This paper further explores the ability of artificial neural networks to estimate three dimensional coordinates from a dynamic stereo head. ...
... Each network was trained with 1000, 1776, 2000, 4000, and 8000 training examples. For each set data was selected both randomly and systematically from approximately 16,000 examples gathered using the method described by Neubert et. al. [2001]. The random selection method occurred without replacement and each sample had the same probablity of being selected. The systematic method attempts to minimize the maximum amount of space between training examples. Because we know the position of the points this is a trivial task. We conjectured that that this method would provide a bet ...
A neural network is used for three-dimensional (3D) reconstruction of a point from a pair of images obtained with an active stereo system. Our active stereo system describes the position of a point with eight parameters: two pan an- gles, two tilt angles, and two-dimensional coordinates of the projected point in each image. Three-dimensional (3D) reconstruction consists of learning the function which maps these eight parameters to the 3D world coordinates (xw, yw, and zw). This paper outlines two possible networks for learning the mapping function. One contains simple Gaussian neurons in the hidden layer. The other is composed of neurons based on a model of the neurons in the parietal cortex of the human brain thought to be involved in 3D re- construction from visual data. The paper compares the performance of each network. We evaluate the size of training set required and the effect of the selection method used to obtain the training examples.
A new technique using the Group Method of Data Handling (GMDH) is proposed for accurate camera calibration. A simple linear pin-hole model is employed first to describe the perspective projection of a camera. Then, the nonlinearity due mainly to lens distortion is learned by the GMDH algorithm. This method provides accurate and reliable results in various camera systems of different optical features unlike most existing techniques based on physical models. Compared to techniques using artificial neural networks (ANNs), which are also capable of learning nonlinear relations without modelling, learning by GMDH is much faster. We prove the advantages of the proposed technique with various simulated data and in real experiments.
BP neural network can solve the complex nonlinear problem, with which it is very difficult for a binocular stereovision to build a complete calibration mathematical model containing all lens distortions and noises. The BP neural network is used for calibrating a binocular stereovision. Furthermore, the effect of the number of network layer on the calibration of the binocular stereovision is studied. Four BP neural networks are constructed with different number of layer. And an experiment is performed to test the calibration capability of the four networks. The test results demonstrate that the network model with three layers has higher calibration efficiency, stronger generalization ability and higher working accuracy, compared with the network models with four, five and six layers. ©, 2014, Editorial Board of Optical Technique. All right reserved.
Camera calibration is an important and fundamental procedure for the application of a vision sensor to 3D problems. Recently many camera calibration methods have been proposed particularly in the area of robot vision. However, the reliability of data used in calibration has been seldomly considered in spite of its importance. In addition, a camera model can not guarantee good results consistently in various conditions. This paper proposes methods to overcome such uncertainty problems of data and camera models as we often encounter them in practical camera calibration steps. By the use of the RANSAC (Random Sample Consensus) algorithm, few data having excessive magnitudes of errors are excluded. Artificial neural networks combined in a two-step structure are trained to compensate for the result by a calibration method of a particular model in a given condition. The proposed methods are useful because they can be employed additionally to most existing camera calibration techniques if needed. We applied them to a linear camera calibration method and could get improved results.
