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.     

DeViouS: A distributed environment for computer vision

Computer vision, owing to the size and complexity of its tasks and its importance to industrial and economic growth, was selected as one of the grand challenge problems by the U.S. Federal High Performance Computing Program. Integration of vision operations is identified as a key element of the challenge. A system to integrate computer vision in a distributed environment is presented here. This system, called DeViouS, is based on the client/server model and runs in a heterogeneous environment of Unix workstations. Modern computing environments include large numbers of high-powered workstations connected by a very fast network. Many of these computers are idle most of the time. DeViouS takes advantage of this feature of computing environments to distribute the execution of vision tasks. Two primary goals of DeViouS are to provide a practical distributed system and a research environment for vision computing. DeViouS is based on a modular design that allows experimentation in various aspects of algorithm design, scheduling and network programming. It can make use of any existing computer vision packages with very minor changes to DeViouS. DeViouS has been tested in an environment of SUN and Digital workstations and has shown substantial improvements in speed over sequential computing with negligible overhead.     

The design of distributed control computer systems

The top-down approach in the design of Distributed Control Computer Systems, DCCS, is presented. The starting point of the study is three statements: (i) the design approach should be production process control oriented; (ii) for real time process control applications, computer networks require an efficient communication subsystem, the most important features of which are high reliability and short message transmission times and (iii) recent research has shown that the tools and methodologies used in both computing and control should be complementary to each other in order to solve some of the distributed control computer system/distributed computer control system design problems. The paper contains the design problems of distributed control computer systems, the presentation of the methods and tools for the study and design of the DCCS, the top-down approach in the design of the DCCS, and examples of two design stages.     

Output feedback hierarchical control for nonlinear systems

In this paper, we address the problem of hierarchical control for nonlinear systems and design two dynamic output feedback hierarchical control laws in a semiglobal sense and in a global sense, respectively. With the controllers applied to a class of nonlinear systems, some transient and steady properties of the system output trajectories can be satisfied simultaneously. Furthermore, the corresponding result in the global sense for linear systems is naturally derived. Finally, the effectiveness of our approach is illustrated by a single‐link robot arm system. Copyright © 2017 John Wiley & Sons, Ltd.     

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.     

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.

一种面向机器视觉感知的暗光图像增强网络

低光照等恶劣环境下的目标检测一直都是难点,低光照和多雾因素往往会导致图像出现可视度低、噪声大等情况,严重干扰目标检测的检测精度。针对上述问题,提出了一个面向机器视觉感知的低光图像增强网络MVP-Net,并与YOLOv3目标检测网络整合,构建了端到端的增强检测框架MVP-YOLO。MVP-Net采用了逆映射网络技术,将常规RGB图像转换为伪RAW图像特征空间,并提出了伪ISP增强网络DOISP进行图像增强。MVP-Net旨在发挥RAW图像在目标检测中的潜在优势,同时克服其在直接应用时所面临的限制。模型在多个真实场景暗光数据上取得了优于先前工作效果并且能够适应多种不同架构的检测器。其端到端检测框mAP（50%）指标达到了78.3%,比YOLO检测器提高了1.85%。     

Characterization of equilibrium sets for bilinear systems with feedback control

Simple analytical methods for the characterization of the equilibrium sets of bilinear systems with linear feedback control are developed based on elimination theory. It is shown that the number of equilibrium points can be determined as a function of the feedback gains, without having to solve for the locations of the equilibrium points. Necessary and sufficient conditions for the origin to be the unique equilibrium point are also presented. These results provide the basis for computing the equilibrium points via a straightforward procedure, which eliminates the convergence, stability and uncertainty problems that can occur when iterative root finding techniques are used. Several illustrative examples are included. The methods developed in this paper can also be used to characterize the equilibrium sets of other classes of nonlinear systems such as quadratic systems.     

智能制造中的计算机视觉应用瓶颈问题

计算机视觉在智能制造工业检测中发挥着检测识别和定位分析的重要作用,为提高工业检测的检测速率和准确率以及智能自动化程度做出了巨大的贡献。然而计算机视觉在应用过程中一直存在技术应用难点,其中3大瓶颈问题是：计算机视觉应用易受光照影响、样本数据难以支持深度学习、先验知识难以加入演化算法。这些瓶颈问题使得计算机视觉在智能制造中的应用无法发挥最佳效能。因此,需要系统地加以分析和解决。本文总结了智能制造和计算机视觉的概念及其重要性,分析了计算机视觉在智能制造工业检测领域的发展现状和需求。针对计算机视觉应用存在的3大瓶颈问题总结分析了问题现状和已有解决方法。经过深入分析发现：针对受光照影响大的问题,可以通过算法和图像采集两个环节解决;针对样本数据难以支持深度学习的问题,可以通过小样本数据处理算法和样本数量分布平衡方法解决;针对先验知识难以加入演化算法的问题,可以通过机器学习和强化学习解决。上述解决方案中的方法不尽相同,各有优劣,需要结合智能制造中具体应用研究和改进。     

Output feedback control of nonlinear systems subject to sensor data losses

In this work, we focus on output feedback control of nonlinear systems subject to sensor data losses. We initially construct an output feedback controller based on a combination of a Lyapunov-based controller with a high-gain observer. We then study the stability and robustness properties of the closed-loop system in the presence of sensor data losses for both the continuous and sampled-data systems. We state a set of sufficient conditions under which the closed-loop system is guaranteed to be practically stable. The theoretical results are demonstrated using a chemical process example.     

A dynamic feedback control strategy for control loops with time-varying delay

Dynamic systems of nth order with time-varying delay in the control loop are examined in this paper. The infinite-dimensional pure delay problem is approximated using a jth-order Padé approximation. Although the approximation provides a well-matched finite-dimensional configuration, it poses a new challenge in terms of unstable internal dynamics for the resulted non-minimum phase system. Such a non-minimum phase characteristic limits the closed-loop system bandwidth and leads to an imperfect tracking performance. To circumvent this problem, the unstable internal dynamics of the system is captured and a new dynamic compensator is proposed to stabilise it in a systematic framework. A dynamic controller is developed, which provides the overall system stability against unmatched perturbation and meets the desired tracking error dynamics. The proposed approach is then applied to fuelling control in gasoline engines addressing the varying transport delay of the oxygen-sensor measurement in the exhaust. The developed methodology is finally validated on a Ford F-150 SI lean-burn engine model with large time-varying delay in the control loop.     

基于快速输出采样反馈的网络控制系统滑模控制

针对网络控制系统的网络诱导时延问题,提出了一种新的滑模控制方法。通过快速采样被控对象输出,建立了具有线性时不变被控对象的网络控制系统的离散模型。在此基础上,用极点配置法给出了滑模切换面参数,设计了输出反馈离散滑模控制器。仿真结果验证了该控制方法的有效性。     

A note to output feedback adaptive control for uncertain system with static nonlinearity

The problem of control design for a system represented as linear stationary and static nonlinear parts is considered. It is assumed that the linear part is unknown and strictly minimum phase. The nonlinear part is known inaccurately, it is irreducible to an input of the linear block, and generally does not satisfy sector restrictions. An adaptive regulator ensuring asymptotic stability is synthesized. The output of a control system, but not its derivatives, is used as a measured variable.     

Robust optimal control during feedback disruption for nonlinear systems controlled by an observer-based output feedback controller

In this paper, a robust optimal control problem during feedback disruption is considered for a class of nonlinear systems which have been controlled by an observer-based output feedback controller. It is shown that during feedback disruption, there exists an optimal control input which keeps both system states and observer errors within a specified bound for the longest time. Then, it is shown that such an optimal control input can be practically implemented by using a bang-bang control input in terms of control performance. One numerical and one practical examples are given for clear illustration.     

Intelligent control of a robot manipulator by visual feedback

We present a new approach to visual feedback control using image-based visual servoing with stereo vision. In order to control the position and orientation of a robot with respect to an object, a new technique is proposed using binocular stereo vision. The stereo vision enables us to calculate an exact image Jacobian not only at around a desired location, but also at other locations. The suggested technique can guide a robot manipulator to the desired location without needing such a priori knowledge as the relative distance to the desired location or a model of the object, even if the initial positioning error is large. We describes a model of stereo vision and how to generate feedback commands. The performance of the proposed visual servoing system is illustrated by a simulation and by experimental results, and compared with the conventional method for an assembling robot. This work was presented in part at the Fourth International Symposium on Artificial Life and Robotics, Oita, Japan, January 19–22, 1999     

Continuous-discrete gain transformation methods for linear feedback control

The problem of transforming between continuoustime state variable feedback gains and equivalent discrete gains suitable for digital implementation is considered. The concepts of state and control equivalence yield two simple transformation rules, a pseudo-inverse method and an average gain method, respectively. As the sampling interval δ→0, these methods are contrasted with existing Taylor series based approaches. The new transformation rules are also studied numerically using a ship course-keeping example. Transformed optimal continuous gains are compared with optimal discrete gains over a wide range of sampling intervals.     

Distributed containment control for nonlinear multiagent systems in pure‐feedback form

In this paper, the problem of distributed containment control for pure‐feedback nonlinear multiagent systems under a directed graph topology is investigated. The dynamics of each agent are molded by high‐order nonaffine pure‐feedback form. Neural networks are employed to identify unknown nonlinear functions, and dynamic surface control technique is used to avoid the problem of explosion of complexity inherent in backstepping design procedure. The Frobenius norm of the ideal neural network weighting matrices is estimated, which is helpful to reduce the number of the adaptive tuning law and alleviate the networked communication burden. The proposed distributed containment controllers guarantee that all signals in the closed‐loop systems are cooperatively semiglobally uniformly ultimately bounded, and the outputs of followers are driven into a convex hull spanned by the multiple dynamic leaders. Finally, the effectiveness of the developed method is demonstrated by simulation examples.     

Quantized output feedback control for networked control systems

This paper addresses the problem of output feedback control for networked control systems (NCSs) with limited communication capacity. Firstly, we propose a new model to describe the non-ideal network conditions and the input/output state quantization of the NCSs in a unified framework. Secondly, based on our newly proposed model and an improved separation lemma, the observer-based controller is developed for the asymptotical stabilization of the NCSs, which are shown in terms of nonlinear matrices inequalities. The nonlinear problems can be computed through solving a convex optimization problems, and the observed and controller gains could be derived by solving a set of linear matrix inequalities. Thirdly, two simulation examples are given to demonstrate the effectiveness of the proposed method.