Parallel computer vision on Polymorphic Torus architecture

Polymorphic Torus is a novel interconnection network for SIMD massively parallel computers, able to support effectively both local and global communication. Thanks to this characteristic, Polymorphic Torus is highly suitable for computer vision applications, since vision involves local communication at the low-level stage and global communication at the intermediate- and high-level stages. In this paper we evaluate the performance of Polymorphic Torus in the computer vision domain. We consider a set of basic vision tasks, namely,convolution, histogramming, connected component labeling, Hough transform, extreme point identification, diameter computation, andvisibility, and show how they can take advantage of the Polymorphic Torus communication capabilities. For each basic vision task we propose a Polymorphic Torus parallel algorithm, give its computational complexity, and compare such a complexity with the complexity of the same task inmesh, tree, pyramid, and hypercube interconnection networks. In spite of the fact that Polymorphic Torus has the same wiring complexity as mesh, the comparison shows that in all of the vision tasks under examination it achieves complexity lower than or at most equal to hypercube, which is the most powerful among the interconnection networks considered.     

Mathematical statistics and computer vision

In this discussion paper, I present my views on the role on mathematical statistics for solving computer vision problems.     

The role of computer vision in prosthetic vision

The cost of vision loss worldwide has been estimated at nearly $3 trillion (http://www.amdalliance.org/cost-of-blindness.html). Non-preventable diseases cause a significant proportion of blindness in developed nations and will become more prevalent as people live longer. Prosthetic vision technologies including retinal implants will play an important therapeutic role. Retinal implants convert an input image stream to visual percepts via stimulation of the retina. This paper highlights some barriers to restoring functional human vision for current generation visual prosthetic devices that computer vision can help overcome. Such computer vision is interactive, aiming to restore function including visuo-motor tasks and recognition.     

Robot guidance using computer vision

This paper describes a procedure for locating an autonomous robot vehicle in a three-dimensional space. The method is an extension of a two-dimensional technique described by Fukui. The method is simple, requires little computation and provides the unique three-dimensional position of the robot relative to a standard mark.     

Agent-based computer vision in a dynamic, real-time environment

For computer vision systems to operate in many real-world environments, processing must occur in real-time under dynamic conditions. An agent-based methodology offers an approach to increase flexibility and scalability to accommodate the demands of a real-time, dynamic environment. This paper presents an agent-based architecture that uses a utility optimization technique to guarantee that important vision tasks are fulfilled even under resource constraints. To ensure that the processing of vision tasks is both reliable and flexible, multiple behaviors are utilized to accomplish the vision application's requirements. A vision behavior consists of a grouping of vision algorithms and a set of service levels associated with these algorithms. Utility functions are adopted to evaluate the performance of all possible behaviors that can address the requirements of a vision application within resource constraints. The maximum overall utility corresponds to the optimal behavior. Two example systems using this model are presented to show the applicability of the architecture. Experimental results show that this agent-based architecture outperforms traditional non-agent-based approaches.     

MinGPU: a minimum GPU library for computer vision

In the field of computer vision, it is becoming increasingly popular to implement algorithms, in sections or in their entirety, on a graphics processing unit (GPU). This is due to the superior speed GPUs offer compared to CPUs. In this paper, we present a GPU library, MinGPU, which contains all of the necessary functions to convert an existing CPU code to GPU. We have created GPU implementations of several well known computer vision algorithms, including the homography transformation between two 3D views. We provide timing charts and show that our MinGPU implementation of homography transformations performs approximately 600 times faster than its C++ CPU implementation.

Detecting corn tassels using computer vision and support vector machines

An automated solution for maize detasseling is very important for maize growers who want to reduce production costs. Quality assurance of maize requires constantly monitoring production fields to ensure that only hybrid seed is produced. To achieve this cross-pollination, tassels of female plants have to be removed for ensuring all the pollen for producing the seed crop comes from the male rows. This removal process is called detasseling. Computer vision methods could help positioning the cutting locations of tassels to achieve a more precise detasseling process in a row. In this study, a computer vision algorithm was developed to detect cutting locations of corn tassels in natural outdoor maize canopy using conventional color images and computer vision with a minimum number of false positives. Proposed algorithm used color informations with a support vector classifier for image binarization. A number of morphological operations were implemented to determine potential tassel locations. Shape and texture features were used to reduce false positives. A hierarchical clustering method was utilized to merge multiple detections for the same tassel and to determine the final locations of tassels. Proposed algorithm performed with a correct detection rate of 81.6% for the test set. Detection of maize tassels in natural canopy images is a quite difficult task due to various backgrounds, different illuminations, occlusions, shadowed regions, and color similarities. The results of the study indicated that detecting cut location of corn tassels is feasible using regular color images.     

Modular architecture for custom-built systems oriented to real-time computer vision: Application to color recognition

This paper presents a modular architecture called DIPSA, which is intended to be used for building custom-made real-time Computer-Vision systems. It consists of four module types and each of them represents a family of circuits that perform specific visual tasks. Our architectural model proposes an algorithm-dependent methodology and makes good results possible using problem oriented solutions. The desired performance is achieved choosing the appropriate modules and connecting them by means of heterogeneous pipeline and concurrence. Additionally, two DIPSA-based hardware systems for real-time Color Recognition are described here.     

Toward designing intelligent PDEs for computer vision: An optimal control approach

Many computer vision and image processing problems can be posed as solving partial differential equations (PDEs). However, designing a PDE system usually requires high mathematical skills and good insight into the problems. In this paper, we consider designing PDEs for various problems arising in computer vision and image processing in a lazy manner: learning PDEs from training data via an optimal control approach. We first propose a general intelligent PDE system which holds the basic translational and rotational invariance rule for most vision problems. By introducing a PDE-constrained optimal control framework, it is possible to use the training data resulting from multiple ways (ground truth, results from other methods, and manual results from humans) to learn PDEs for different computer vision tasks. The proposed optimal control based training framework aims at learning a PDE-based regressor to approximate the unknown (and usually nonlinear) mapping of different vision tasks. The experimental results show that the learnt PDEs can solve different vision problems reasonably well. In particular, we can obtain PDEs not only for problems that traditional PDEs work well but also for problems that PDE-based methods have never been tried before, due to the difficulty in describing those problems in a mathematical way.     

Towards automated progress assessment of workpackage components in construction projects using computer vision

Traditional methods of collecting data on the progress of construction projects often involve human judgement, high costs, and are too infrequent to provide managers with timely and accurate control data. This study proposes a system that aims to address some of these shortcomings through automation. Our first contribution is a framework for the automatic generation of work packages. We describe the conceptual model and develop a prototype implementation, based on the results of a survey designed to capture the current practice of formulating work packages as performed in the industry. Our second contribution is a system that employs computer vision (CV) techniques to report on the progress of these work packages automatically. This is achieved by analysing sequences of digital images acquired regularly by on-site cameras. The result of this analysis, when compared with the expected state of construction, provides progress feedback on the work packages, allowing a project to be monitored more effectively.     

Uncalibrated reconstruction: an adaptation to structured light vision

Euclidean reconstruction from two uncalibrated stereoscopic views is achievable from the knowledge of geometrical constraints about the environment. Unfortunately, these constraints may be quite difficult to obtain. In this paper, we propose an approach based on structured lighting, which has the advantage of providing geometrical constraints independent of the scene geometry. Moreover, the use of structured light provides a unique solution to the tricky correspondence problem present in stereovision. The projection matrices are first computed by using a canonical representation, and a projective reconstruction is performed. Then, several constraints are generated from the image analysis and the projective reconstruction is upgraded into an Euclidean one—as we will demonstrate, it is assumed that the sensor behaviour is affine without loss of generality so that the constraints generation is simplified. The method provides our sensor with adaptive capabilities and permits to be used in the measurement of moving scenes such as dynamic visual inspection or mobile robot navigation. Experimental results obtained from both simulated and real data are presented.     

Accurate and speedy computation of image Legendre moments for computer vision applications

A novel algorithm that permits the fast and accurate computation of the Legendre image moments is introduced in this paper. The proposed algorithm is based on the block representation of an image and on a new image representation scheme, the Image Slice Representation (ISR) method. The ISR method decomposes a gray-scale image as an expansion of several two-level images of different intensities (slices) and thus enables the partial application of the well-known Image Block Representation (IBR) algorithm to each image component. Moreover, using the resulted set of image blocks, the Legendre moments’ computation can be accelerated through appropriate computation schemes. Extensive experiments prove that the proposed methodology exhibits high efficiency in calculating Legendre moments on gray-scale, but furthermore on binary images. The newly introduced algorithm is suitable for the computation of the Legendre moments for pattern recognition and computer vision applications, where the images consist of objects presented in a scene.     

Context aided pedestrian detection for danger estimation based on laser scanner and computer vision

Road safety applications demand the most reliable sensor systems. In recent years, the advances in information technologies have led to more complex road safety applications able to cope with a high variety of situations. These applications have strong sensing requirements that a single sensor, with the available technology, cannot attain. Recent researches in Intelligent Transport Systems (ITS) try to overcome the limitations of the sensors by combining them. But not only sensor information is crucial to give a good and robust representation of the road environment; context information has a key role for reliable safety applications to provide reliable detection and complete situation assessment. This paper presents a novel approach for pedestrian detection using sensor fusion of laser scanner and computer vision. The application also takes advantage of context information, providing danger estimation for the pedestrians detected. Closing the loop, the danger estimation is later used, together with context information, as feedback to enhance the pedestrian detection process.     

Line segment matching for 3D computer vision using a new iteration scheme

A new iteration scheme is proposed to solve the line segment matching problem in stereo vision analysis. A match function which directly reflects the requirements of the epipolar and disparity constraints is proposed for line segment matching. The information contained in the match function is used to determine line segment correspondences indirectly. After a match network is established according to the match function values, a new iteration algorithm is employed to tune the strengths of the match links in the match network so that the match network can converge to a stable state. No explicit compatibility coefficient need be defined for computing the support function values in the iterations, resulting in a faster computation speed than those of conventional relaxation matching techniques. The inherent anti-symmetric characteristic of relaxation matching for the image correspondence problem is also avoided naturally. The experimental results show that the proposed iteration scheme is effective and suitable for matching line segments even when images are complicated.     

双目立体匹配算法的研究与进展

立体匹配一直是计算机视觉领域的一个中心研究问题.首先综合介绍了立体匹配算法的研究概况,论述了双目立体匹配算法中各种约束的核心概念和适用范围;然后重点归纳分析了立体匹配算法的分类及其发展过程中的各种演化算法,对其关键技术进行了剖析和比较,并总结了目前存在的主要难题和可能的解决途径;最后对该领域存在的问题和技术发展趋势进行了分析和讨论.     

Theory of edge perception for computer vision feedback control

In this paper, the authors address the problem of edge-perception for its applications to vision-feedback control in robotic systems.In natural vision, the recognition of objects takes place through the process consisting of eye system, neural networks and cognition. The cognitive process, in turn yields a phenomenon known as perception. This is the phenomenon of perception of physical attributes, such as edges, color and texture, etc., which is responsible for the recognition of objects through the natural vision processes.In this paper, we make an attempt to postulate the theory of perception for gray-level images. The gray-level images, when going through the cognitive and perception processes, are contaminated by the uncertainty; here we call it cognitive uncertainty.The studies in this paper are confined to the phenomenon of edge-perception for two-dimensional gray-level images, however, these studies can be extended to other types of visual attributes both in two-dimensional and three-dimensional spaces. Indeed, the perception of these attributes, which attempts to emulate the human vision system, may help in the design of a truly robust computer vision-feedback control system for robotic applications.     

Novel moment invariants for improved classification performance in computer vision applications

A novel set of moment invariants based on the Krawtchouk moments are introduced in this paper. These moment invariants are computed over a finite number of image intensity slices, extracted by applying an innovative image representation scheme, the image slice representation (ISR) method. Based on this technique an image is decomposed to a several non-overlapped intensity slices, which can be considered as binary slices of certain intensity. This image representation gives the advantage to accelerate the computation of image's moments since the image can be described in a number of homogenous rectangular blocks, which permits the simplification of the computation formulas. The moments computed over the extracted slices seem to be more efficient than the corresponding moments of the same order that describe the whole image, in recognizing the pattern under processing. The proposed moment invariants are exhaustively tested in several well known computer vision datasets, regarding their rotation, scaling and translation (RST) invariant recognition performance, by resulting to remarkable outcomes.     

The least-disturbance principle and weak constraints

Certain problems, notably in computer vision, involve adjusting a set of real-valued labels to satisfy certain constraints. They can be formulated as optimisation problems, using the ‘least-disturbance’ principle: the minimal alteration is made to the labels that will achieve a consistent labelling. Under certain linear constraints, the solution can be achieved iteratively and in parallel, by hill-climbing. However, where ‘weak’ constraints are imposed on the labels — constraints that may be broken at a cost — the optimisation problem becomes non-convex; a continuous search for the solution is no longer satisfactory. A strategy is proposed for this case, by construction of convex envelopes and by the use of ‘graduated’ non-convexity.     

3-D shape reconstruction in an active stereo vision system using genetic algorithms

The recovery of 3-D shape information (depth) using stereo vision analysis is one of the major areas in computer vision and has given rise to a great deal of literature in the recent past. The widely known stereo vision methods are the passive stereo vision approaches that use two cameras. Obtaining 3-D information involves the identification of the corresponding 2-D points between left and right images. Most existing methods tackle this matching task from singular points, i.e. finding points in both image planes with more or less the same neighborhood characteristics. One key problem we have to solve is that we are on the first instance unable to know a priori whether a point in the first image has a correspondence or not due to surface occlusion or simply because it has been projected out of the scope of the second camera. This makes the matching process very difficult and imposes a need of an a posteriori stage to remove false matching.In this paper we are concerned with the active stereo vision systems which offer an alternative to the passive stereo vision systems. In our system, a light projector that illuminates objects to be analyzed by a pyramid-shaped laser beam replaces one of the two cameras. The projections of laser rays on the objects are detected as spots in the image. In this particular case, only one image needs to be treated, and the stereo matching problem boils down to associating the laser rays and their corresponding real spots in the 2-D image. We have expressed this problem as a minimization of a global function that we propose to perform using Genetic Algorithms (GAs). We have implemented two different algorithms: in the first, GAs are performed after a deterministic search. In the second, data is partitioned into clusters and GAs are independently applied in each cluster. In our second contribution in this paper, we have described an efficient system calibration method. Experimental results are presented to illustrate the feasibility of our approach. The proposed method yields high accuracy 3-D reconstruction even for complex objects. We conclude that GAs can effectively be applied to this matching problem.